# Corey James

> AI systems architect and research engineer. Original systems in GPU inference, inference memory, autonomous agent governance and enterprise AI, built and measured on hardware the author owns. Every page states the problem, the decisions and the measured result.

Scale: 4.04 M lines of source, 5,048 commits, 12 languages, 10 systems, 90 engineering stories, 15 ideas, 517 MCP tools, 5 provisional patent applications.

Machine-readable inventory: https://coreyjames.ai/agent-index.json
MCP server (streamable HTTP; match_role gives evidence-cited fit reports): https://coreyjames.ai/mcp
HTTP API for agents that can only browse (GET /api/<tool>?arg=value, add format=md for Markdown; OpenAPI https://coreyjames.ai/openapi.json): https://coreyjames.ai/api/match_role?requirements=..., https://coreyjames.ai/api/search_portfolio?query=..., https://coreyjames.ai/api/get_page?url=..., https://coreyjames.ai/api/who
Manifest: https://coreyjames.ai/.well-known/mcp.json
A2A agent (message/send: fit report, search, who, contact): https://coreyjames.ai/a2a, card https://coreyjames.ai/.well-known/agent.json, skills https://coreyjames.ai/.well-known/skills/index.json
Contact: the form at https://coreyjames.ai/about/#contact, or the contact tool on the MCP server and the A2A agent (sender's own name and email required)
Contact: https://coreyjames.ai/about/#contact
Repository documents of every page (GitHub): https://github.com/coreyjames-ai
Repository documents of every page (GitLab mirror): https://gitlab.com/coreyjames-ai
ORCID: https://orcid.org/0009-0008-7246-963X

## Repositories
Each repository holds the same documents as the site section it mirrors: every page as a README with its measured figures, evidence lines and canonical link, plus the page record as JSON.
- activekv: An inference-memory layer for vLLM: the conversation's history in host RAM, the GPU as a bounded working set · KV-cache offload for LLM inference, Intel GPU and SYCL/oneAPI engineering, C++ and native ports with parity proof, GPU inference optimisation · GitHub https://github.com/coreyjames-ai/activekv · GitLab https://gitlab.com/coreyjames-ai/activekv · DOI https://doi.org/10.5281/zenodo.23735116 · page https://coreyjames.ai/systems/activekv/
- agent-swarm: A self-hosted, multi-host runtime where tools are the platform and agents are compositions · sandboxing and container hardening, agent orchestration · GitHub https://github.com/coreyjames-ai/agent-swarm · GitLab https://gitlab.com/coreyjames-ai/agent-swarm · DOI https://doi.org/10.5281/zenodo.23735118 · page https://coreyjames.ai/systems/agent-swarm/
- agentos: A governed work-order control plane for teams of AI engineering agents · cost attribution for AI work, C++ and native ports with parity proof, evidence-gated acceptance, separation of duties for AI work · GitHub https://github.com/coreyjames-ai/agentos · GitLab https://gitlab.com/coreyjames-ai/agentos · DOI https://doi.org/10.5281/zenodo.23735120 · page https://coreyjames.ai/systems/agentos/
- auth-and-access-plane: How commercial AI assistants and my own agents reach a private tool estate: OAuth, token validation, edge exposure and an operations layer · OAuth 2.0 and identity architecture, zero-trust and zero-inbound networking · GitHub https://github.com/coreyjames-ai/auth-and-access-plane · GitLab https://gitlab.com/coreyjames-ai/auth-and-access-plane · DOI https://doi.org/10.5281/zenodo.23735122 · page https://coreyjames.ai/systems/auth-and-access-plane/
- capabilities-model: A 0.6-billion-parameter model that memorises an agent's tool catalog, with the data, decoder, serving and evaluation around it · TensorRT and TensorRT-LLM engineering, inference serving architecture, CUDA kernel development, MCP server design · GitHub https://github.com/coreyjames-ai/capabilities-model · GitLab https://gitlab.com/coreyjames-ai/capabilities-model · DOI https://doi.org/10.5281/zenodo.23735124 · page https://coreyjames.ai/systems/capabilities-model/
- coreyjames-ai: The profile · GitHub https://github.com/coreyjames-ai/coreyjames-ai · GitLab https://gitlab.com/coreyjames-ai/coreyjames-ai · page https://coreyjames.ai/
- fafo-core: The execution substrate under everything else: one function definition becomes an orchestrated, budgeted, agent-callable tool · exactly-once delivery and crash recovery, tool synthesis and governed API integration, high-throughput batch inference, PostgreSQL performance tuning · GitHub https://github.com/coreyjames-ai/fafo-core · GitLab https://gitlab.com/coreyjames-ai/fafo-core · DOI https://doi.org/10.5281/zenodo.23735130 · page https://coreyjames.ai/systems/fafo-core/
- fafohive: The MCP hub every agent in my estate connects to: memory and code intelligence, hosted tools, and a governed tool gateway · exactly-once delivery and crash recovery, MCP server design · GitHub https://github.com/coreyjames-ai/fafohive · GitLab https://gitlab.com/coreyjames-ai/fafohive · DOI https://doi.org/10.5281/zenodo.23735132 · page https://coreyjames.ai/systems/fafohive/
- helix: Humans and several AI minds in one room, and one colleague who can be reached wherever the work is: in team chat, by voice, and in a Teams call · exactly-once delivery and crash recovery, Microsoft Teams real-time media, agent orchestration, real-time voice AI · GitHub https://github.com/coreyjames-ai/helix · GitLab https://gitlab.com/coreyjames-ai/helix · DOI https://doi.org/10.5281/zenodo.23735146 · page https://coreyjames.ai/systems/helix/
- inference-fabric: A throughput-first GPU inference pipeline on consumer NVIDIA hardware: broker, saturated runtime, completion, vector store · high-throughput batch inference, token-aware scheduling, GPU inference optimisation, CUDA kernel development · GitHub https://github.com/coreyjames-ai/inference-fabric · GitLab https://gitlab.com/coreyjames-ai/inference-fabric · DOI https://doi.org/10.5281/zenodo.23735148 · page https://coreyjames.ai/systems/inference-fabric/
- portfolio: Corey James: the portfolio as citable research. 10 systems, 90 engineering stories, 5 provisional patent applications, every figure on the page it cites. · GitHub https://github.com/coreyjames-ai/portfolio · GitLab https://gitlab.com/coreyjames-ai/portfolio · DOI https://doi.org/10.5281/zenodo.23735152 · page https://coreyjames.ai/systems/
- postgresql-fabric: One governed write path for a GPU-scale estate: batches sized from the database's own limits, workers that follow the queue, admission that reads the database's pressure, and the heavy work run inside · PostgreSQL performance tuning, Rust systems engineering, database admission control, message broker and queue design · GitHub https://github.com/coreyjames-ai/postgresql-fabric · GitLab https://gitlab.com/coreyjames-ai/postgresql-fabric · DOI https://doi.org/10.5281/zenodo.23735156 · page https://coreyjames.ai/systems/postgresql-fabric/

## Systems
- [AgentOS](https://coreyjames.ai/systems/agentos/): A governed work-order control plane for teams of AI engineering agents
- [Agent Swarm](https://coreyjames.ai/systems/agent-swarm/): A self-hosted, multi-host runtime where tools are the platform and agents are compositions
- [Helix](https://coreyjames.ai/systems/helix/): Humans and several AI minds in one room, and one colleague who can be reached wherever the work is: in team chat, by voice, and in a Teams call
- [FAFOHive](https://coreyjames.ai/systems/fafohive/): The MCP hub every agent in my estate connects to: memory and code intelligence, hosted tools, and a governed tool gateway
- [FAFO Core and Function Forge](https://coreyjames.ai/systems/fafo-core/): The execution substrate under everything else: one function definition becomes an orchestrated, budgeted, agent-callable tool
- [PostgreSQL Fabric](https://coreyjames.ai/systems/postgresql-fabric/): One governed write path for a GPU-scale estate: batches sized from the database's own limits, workers that follow the queue, admission that reads the database's pressure, and the heavy work run inside the database next to its indexes
- [Inference Fabric](https://coreyjames.ai/systems/inference-fabric/): A throughput-first GPU inference pipeline on consumer NVIDIA hardware: broker, saturated runtime, completion, vector store
- [ActiveKV](https://coreyjames.ai/systems/activekv/): An inference-memory layer for vLLM: the conversation's history in host RAM, the GPU as a bounded working set
- [Capabilities model](https://coreyjames.ai/systems/capabilities-model/): A 0.6-billion-parameter model that memorises an agent's tool catalog, with the data, decoder, serving and evaluation around it
- [Auth and access plane](https://coreyjames.ai/systems/auth-and-access-plane/): How commercial AI assistants and my own agents reach a private tool estate: OAuth, token validation, edge exposure and an operations layer

## Capabilities
- [Governed autonomous engineering](https://coreyjames.ai/capabilities/governed-autonomous-engineering/): Programs of AI coding agents that finish, prove it, and can be trusted
- [Bigger models and longer context on fixed GPU memory](https://coreyjames.ai/capabilities/longer-context-on-the-gpus-you-own/): When GPU memory, not the workload, is the limit
- [AI teammates inside Microsoft Teams and team chat](https://coreyjames.ai/capabilities/ai-teammates-in-teams-and-chat/): A colleague people can call, interrupt and invite to meetings, under an identity the organisation governs
- [Business systems as safe agent tools](https://coreyjames.ai/capabilities/business-systems-as-safe-agent-tools/): Agents acting on payments, email, social platforms and private APIs without holding a key, with every action attributed
- [High-throughput batch inference and embeddings](https://coreyjames.ai/capabilities/high-throughput-batch-inference/): Millions of items through GPUs I own, with the hardware kept busy and the accounting balanced
- [Custom small models for tool calling and routing](https://coreyjames.ai/capabilities/small-models-for-tool-calling-and-routing/): When a 0.6B model that cannot be wrong in the dangerous way beats a large model reading a catalog
- [Memory and code intelligence for agents](https://coreyjames.ai/capabilities/memory-and-code-intelligence-for-agents/): Agents that remember across sessions and explain code without inventing it
- [Voice agents in calls and meetings](https://coreyjames.ai/capabilities/voice-agents-in-meetings/): Speech recognition, synthesis and turn-taking on my own hardware, joined to the platforms people already use
- [GPU vector search and corpus analytics](https://coreyjames.ai/capabilities/gpu-vector-search-and-corpus-analytics/): Vectors that stay out of the database yet verifiably tied to it, searched and scored on the GPU
- [AI subscriptions as governed workers](https://coreyjames.ai/capabilities/use-the-ai-subscriptions-you-already-have/): Running vendors' authenticated command-line tools as governed workers, alongside API-keyed and local models

## Ideas
- [Disconnected producers and consumers](https://coreyjames.ai/ideas/disconnected-producers-and-consumers/): Nothing waits on anything else: the method under most of my high-throughput work
- [Authority outside the model](https://coreyjames.ai/ideas/authority-outside-the-model/): The model is a nondeterministic worker inside a deterministic system that decides what it may do, what step it is on, and whether it is finished
- [The copy of record lives off the accelerator](https://coreyjames.ai/ideas/the-copy-of-record-lives-off-the-accelerator/): Host memory or the database holds the truth; the GPU or the model holds a bounded working set
- [The token as the unit](https://coreyjames.ai/ideas/the-token-as-the-unit/): Schedule, budget and account for GPU work in the unit the hardware actually pays in
- [Tools are the platform](https://coreyjames.ai/ideas/tools-are-the-platform/): Build the safety machinery once, make it the only path, and let everything else be a declaration
- [Identity from the caller, credentials on the server](https://coreyjames.ai/ideas/identity-from-the-caller/): An agent is who it authenticated as, never who it says it is, and it never holds a key it does not need
- [Compute first, narrate second](https://coreyjames.ai/ideas/compute-first-narrate-second/): Put deterministic evidence in front of the model instead of asking the model for evidence
- [Measure the instrument](https://coreyjames.ai/ideas/measure-the-instrument/): A result is only as good as the proof that the measurement measured the thing
- [Budgets and refusals](https://coreyjames.ai/ideas/budgets-and-refusals/): Spend, memory and admission are first-class controls, and the system says no with a number instead of degrading
- [Reconcile, never resubmit](https://coreyjames.ai/ideas/reconcile-never-resubmit/): When a side effect may or may not have happened, write down the uncertainty and check reality before doing anything again
- [A human at the privileged step](https://coreyjames.ai/ideas/a-human-at-the-privileged-step/): Automate everything up to the one action that carries real authority, and put a person exactly there
- [Role contracts, with vendors as data](https://coreyjames.ai/ideas/role-contracts-not-vendors/): Work is assigned to a role with a contract; which model fills the seat is configuration, and reviewers can be a different vendor from builders on purpose
- [Separate planes by role](https://coreyjames.ai/ideas/separate-planes-by-role/): Traffic with different reliability, latency and durability needs gets its own infrastructure, so pressure in one role cannot starve or corrupt another
- [Append-only, with tombstones and lineage](https://coreyjames.ai/ideas/append-only-with-lineage/): Durable logs are only ever appended, deletion is a mark, and every derived artefact can prove where it came from
- [Agents that react to the world](https://coreyjames.ai/ideas/agents-that-react-to-the-world/): Mailboxes, inbound event sources and standing work orders, so an agent team runs on what happens and never only on prompts

## Patent applications
- [Family Five: Verifiable AI action, tool synthesis and governance](https://coreyjames.ai/research/family-five-verifiable-ai-action-and-tool-synthesis/): Every action is authorised, every output is observed, every result is verifiable; AI cannot self-authorise a tool.
- [Family Four: Self-verifying vector and GPU retrieval substrate](https://coreyjames.ai/research/family-four-self-verifying-vector-substrate/): Same data, same result, verifiable at every step: every index proves it was built from the database it serves.
- [Family One: Deterministic external governance of autonomous AI work](https://coreyjames.ai/research/family-one-deterministic-external-governance/): The AI never decides what it is allowed to do, what step it is on, or whether it is finished. External deterministic rules do.
- [Family Three: Unified context across AI models and knowledge systems](https://coreyjames.ai/research/family-three-unified-context-across-models/): One authoritative context. Many models. Many sources. No AI provider owns the truth.
- [Family Two: Durable work, tombstones and lossless recovery](https://coreyjames.ai/research/family-two-durable-work-and-lossless-recovery/): No accepted work is lost, and no completed result is delivered twice, across a durable broker and GPU runtimes.

## Engineering stories, by the problem solved
- More from consumer GPUs (23): [Context beyond VRAM](https://coreyjames.ai/stories/context-beyond-vram/); [From 44K to 163K tokens per second](https://coreyjames.ai/stories/from-44k-to-163k-tokens-per-second/); [A message broker that understands tokens](https://coreyjames.ai/stories/a-message-broker-that-understands-tokens/); [The workload that shaped the platform](https://coreyjames.ai/stories/the-workload-that-shaped-the-platform/); [Judge a representation by the bytes it stops moving](https://coreyjames.ai/stories/judge-a-representation-by-the-bytes-it-stops-moving/); [Bringing a serving stack up on new silicon](https://coreyjames.ai/stories/bringing-a-serving-stack-up-on-new-silicon/); [A 27B model on one 32 GB Intel card](https://coreyjames.ai/stories/a-27b-model-on-one-32gb-intel-card/); [Spend structure before spending waste](https://coreyjames.ai/stories/spend-structure-before-spending-waste/); [Half the bytes, exactly merged](https://coreyjames.ai/stories/half-the-bytes-exactly-merged/); [Staging without a stall](https://coreyjames.ai/stories/staging-without-a-stall/); [Built for Blackwell](https://coreyjames.ai/stories/built-for-blackwell/); [One call to a GPU anywhere](https://coreyjames.ai/stories/one-call-to-a-gpu-anywhere/); [Engines for the card you have](https://coreyjames.ai/stories/engines-for-the-card-you-have/); [Shapes have a lifecycle](https://coreyjames.ai/stories/shapes-have-a-lifecycle/); [The 128-lane wall](https://coreyjames.ai/stories/the-128-lane-wall/); [Starting full and ending clean](https://coreyjames.ai/stories/starting-full-and-ending-clean/); [Throughput that finished nothing](https://coreyjames.ai/stories/throughput-that-finished-nothing/); [A second engine behind the same door](https://coreyjames.ai/stories/a-second-engine-behind-the-same-door/); [The queue that should not exist](https://coreyjames.ai/stories/the-queue-that-should-not-exist/); [A doorbell the GPU waits on](https://coreyjames.ai/stories/a-doorbell-the-gpu-waits-on/); [The recovery that drained the card](https://coreyjames.ai/stories/the-recovery-that-drained-the-card/); [Prove the mechanism before the model](https://coreyjames.ai/stories/prove-the-mechanism-before-the-model/); [Reading a file format by correlation](https://coreyjames.ai/stories/reading-a-file-format-by-correlation/)
- Proving a system is what it claims (13): [Rewriting an engine extension without changing a token](https://coreyjames.ai/stories/rewriting-an-engine-extension-without-changing-a-token/); [The baseline that tied us](https://coreyjames.ai/stories/the-baseline-that-tied-us/); [Proving a rewrite is the same system](https://coreyjames.ai/stories/proving-a-rewrite-is-the-same-system/); [Bit-identical across batch compositions](https://coreyjames.ai/stories/bit-identical-across-batch-compositions/); [The observer that changed the answer](https://coreyjames.ai/stories/the-observer-that-changed-the-answer/); [Trying the industrial path first](https://coreyjames.ai/stories/trying-the-industrial-path-first/); [A detector for a silent corruption](https://coreyjames.ai/stories/a-detector-for-a-silent-corruption/); [Every metric gets a CPU twin](https://coreyjames.ai/stories/every-metric-gets-a-cpu-twin/); [Asking the engine what it meant](https://coreyjames.ai/stories/asking-the-engine-what-it-meant/); [The thirteen percent that was never missing](https://coreyjames.ai/stories/the-thirteen-percent-that-was-never-missing/); [Tests that passed without a GPU](https://coreyjames.ai/stories/tests-that-passed-without-a-gpu/); [Machinery I built and then did not need](https://coreyjames.ai/stories/machinery-i-built-and-then-did-not-need/); [A proof that proved nothing](https://coreyjames.ai/stories/a-proof-that-proved-nothing/)
- Data paths that cannot lose a row (15): [Millions of queued items that were not there](https://coreyjames.ai/stories/millions-of-queued-items-that-were-not-there/); [196 deadlocks to zero](https://coreyjames.ai/stories/196-deadlocks-to-zero/); [Redistribution inside the database](https://coreyjames.ai/stories/redistribution-inside-the-database/); [Done means delivered](https://coreyjames.ai/stories/done-means-delivered/); [Recovery that trusts only what happened](https://coreyjames.ai/stories/recovery-that-trusts-only-what-happened/); [Twenty thousand rows a second](https://coreyjames.ai/stories/twenty-thousand-rows-a-second/); [A write path that sizes itself](https://coreyjames.ai/stories/a-write-path-that-sizes-itself/); [Rules that never ran](https://coreyjames.ai/stories/rules-that-never-ran/); [Pricing writes by what they cost the database](https://coreyjames.ai/stories/pricing-writes-by-what-they-cost/); [Ghost bytes in a zero-copy ring](https://coreyjames.ai/stories/ghost-bytes-in-a-zero-copy-ring/); [An index you can rebuild from its receipt](https://coreyjames.ai/stories/an-index-you-can-rebuild-from-its-receipt/); [The database decides when an index goes stale](https://coreyjames.ai/stories/the-database-decides-when-an-index-goes-stale/); [Rebuild only what moved](https://coreyjames.ai/stories/rebuild-only-what-moved/); [Three shards that move together](https://coreyjames.ai/stories/three-shards-that-move-together/); [Storage that refuses to become a coordinator](https://coreyjames.ai/stories/storage-that-refuses-to-become-a-coordinator/)
- Measuring embeddings on the GPU (7): [Thirty-eight metrics in a dozen passes](https://coreyjames.ai/stories/thirty-eight-metrics-in-a-dozen-passes/); [Removing the sampling caps](https://coreyjames.ai/stories/removing-the-sampling-caps/); [Where a cluster sits](https://coreyjames.ai/stories/where-a-cluster-sits/); [Diversity in three kernels](https://coreyjames.ai/stories/diversity-in-three-kernels/); [Physics as a lens on embeddings](https://coreyjames.ai/stories/physics-as-a-lens-on-embeddings/); [Two-means as a measuring instrument](https://coreyjames.ai/stories/two-means-as-a-measuring-instrument/); [From health scores to decisions](https://coreyjames.ai/stories/from-health-scores-to-decisions/)
- Keeping AI agents honest (12): [Why long agent runs fail](https://coreyjames.ai/stories/why-long-agent-runs-fail/); [Directing AI coding agents at program scale](https://coreyjames.ai/stories/directing-ai-coding-agents-at-program-scale/); [Closed by evidence](https://coreyjames.ai/stories/closed-by-evidence/); [Agents with real tools and no keys](https://coreyjames.ai/stories/agents-with-real-tools-and-no-keys/); [Separation of duties for AI work](https://coreyjames.ai/stories/separation-of-duties-for-ai-work/); [Tools are the platform](https://coreyjames.ai/stories/tools-are-the-platform/); [A coding agent that never holds the keys](https://coreyjames.ai/stories/a-coding-agent-that-never-holds-the-keys/); [A code explainer that is not allowed to invent](https://coreyjames.ai/stories/a-code-explainer-that-is-not-allowed-to-invent/); [The hardening fix that locked everyone out](https://coreyjames.ai/stories/the-hardening-fix-that-locked-everyone-out/); [Tools that build tools, under governance](https://coreyjames.ai/stories/tools-that-build-tools-under-governance/); [What the work cost, to the penny](https://coreyjames.ai/stories/what-the-work-cost/); [Two thousand concurrent workers](https://coreyjames.ai/stories/two-thousand-concurrent-workers/)
- Small models that constrain big ones (5): [A hundred tokens instead of forty-two thousand](https://coreyjames.ai/stories/a-hundred-tokens-instead-of-forty-two-thousand/); [A router beside a production model](https://coreyjames.ai/stories/a-router-beside-a-production-model/); [Constraint made cheaper than no constraint](https://coreyjames.ai/stories/constraint-made-cheaper-than-no-constraint/); [Every request accounted for](https://coreyjames.ai/stories/every-request-accounted-for/); [A tool catalogue bigger than the model](https://coreyjames.ai/stories/a-tool-catalogue-bigger-than-the-model/)
- Voice that behaves like a person (9): [A self-hosted colleague in a Teams call](https://coreyjames.ai/stories/a-self-hosted-colleague-in-a-teams-call/); [Fast transcripts that were wrong](https://coreyjames.ai/stories/fast-transcripts-that-were-wrong/); [From two seconds to under one](https://coreyjames.ai/stories/from-two-seconds-to-under-one/); [Building the streaming my speech model did not ship](https://coreyjames.ai/stories/building-the-streaming-my-speech-model-did-not-ship/); [Three hundred milliseconds of hello](https://coreyjames.ai/stories/three-hundred-milliseconds-of-hello/); [Saying numbers out loud](https://coreyjames.ai/stories/saying-numbers-out-loud/); [Interruptions that only people can make](https://coreyjames.ai/stories/interruptions-that-only-people-can-make/); [Listen only to me](https://coreyjames.ai/stories/listen-only-to-me/); [One percent network, seventy-eight percent GPU](https://coreyjames.ai/stories/one-percent-network-seventy-eight-percent-gpu/)
- Humans and models in one room (6): [Solving the digital-employee identity](https://coreyjames.ai/stories/solving-the-digital-employee-identity/); [A meeting attendee that will not fake it](https://coreyjames.ai/stories/a-meeting-attendee-that-will-not-fake-it/); [Several minds in one room](https://coreyjames.ai/stories/several-minds-in-one-room/); [Three councils I threw away](https://coreyjames.ai/stories/three-councils-i-threw-away/); [The same brain, typed or spoken](https://coreyjames.ai/stories/the-same-brain-typed-or-spoken/); [From a conversation to a work order](https://coreyjames.ai/stories/from-a-conversation-to-a-work-order/)

## Skills, with the pages that demonstrate each (decided at build time against every page)
- GPU inference optimisation: [From 44K to 163K tokens per second](https://coreyjames.ai/stories/from-44k-to-163k-tokens-per-second/); [GPU inference engine engineering](https://coreyjames.ai/technologies/gpu-inference-engine-engineering/); [High-throughput batch inference and embeddings](https://coreyjames.ai/capabilities/high-throughput-batch-inference/); [Inference Fabric](https://coreyjames.ai/systems/inference-fabric/); [Engines for the card you have](https://coreyjames.ai/stories/engines-for-the-card-you-have/)
- CUDA kernel development: [CUDA C++](https://coreyjames.ai/technologies/cuda-c-plus-plus/); [Thirty-eight metrics in a dozen passes](https://coreyjames.ai/stories/thirty-eight-metrics-in-a-dozen-passes/); [Built for Blackwell](https://coreyjames.ai/stories/built-for-blackwell/); [GPU numerical kernel design](https://coreyjames.ai/technologies/gpu-numerical-kernel-design/); [Tiled pairwise kernels](https://coreyjames.ai/technologies/tiled-pairwise-kernels/)
- TensorRT and TensorRT-LLM engineering: [Inference Fabric](https://coreyjames.ai/systems/inference-fabric/); [From 44K to 163K tokens per second](https://coreyjames.ai/stories/from-44k-to-163k-tokens-per-second/); [NVIDIA TensorRT](https://coreyjames.ai/technologies/nvidia-tensorrt/); [Trying the industrial path first](https://coreyjames.ai/stories/trying-the-industrial-path-first/); [NVIDIA TensorRT-LLM](https://coreyjames.ai/technologies/nvidia-tensorrt-llm/)
- KV-cache offload for LLM inference: [KV-cache engineering](https://coreyjames.ai/technologies/kv-cache-engineering/); [Context beyond VRAM](https://coreyjames.ai/stories/context-beyond-vram/); [vLLM](https://coreyjames.ai/technologies/vllm/); [The copy of record lives off the accelerator](https://coreyjames.ai/ideas/the-copy-of-record-lives-off-the-accelerator/); [GPU memory management](https://coreyjames.ai/technologies/gpu-memory-management/)
- LLM quantization (FP8, INT8, 4-bit): [Model quantization](https://coreyjames.ai/technologies/model-quantization/); [Model families and types](https://coreyjames.ai/technologies/model-families-and-types/); [Hugging Face Hub](https://coreyjames.ai/technologies/hugging-face-hub/); [Bigger models and longer context on fixed GPU memory](https://coreyjames.ai/capabilities/longer-context-on-the-gpus-you-own/); [A 27B model on one 32 GB Intel card](https://coreyjames.ai/stories/a-27b-model-on-one-32gb-intel-card/)
- inference serving architecture: [Capabilities model](https://coreyjames.ai/systems/capabilities-model/); [A router beside a production model](https://coreyjames.ai/stories/a-router-beside-a-production-model/); [A message broker that understands tokens](https://coreyjames.ai/stories/a-message-broker-that-understands-tokens/); [The copy of record lives off the accelerator](https://coreyjames.ai/ideas/the-copy-of-record-lives-off-the-accelerator/); [NVIDIA TensorRT-LLM](https://coreyjames.ai/technologies/nvidia-tensorrt-llm/)
- vLLM internals: [vLLM](https://coreyjames.ai/technologies/vllm/); [Context beyond VRAM](https://coreyjames.ai/stories/context-beyond-vram/); [Rewriting an engine extension without changing a token](https://coreyjames.ai/stories/rewriting-an-engine-extension-without-changing-a-token/); [KV-cache engineering](https://coreyjames.ai/technologies/kv-cache-engineering/); [pybind11](https://coreyjames.ai/technologies/pybind11/)
- Intel GPU and SYCL/oneAPI engineering: [Intel oneAPI](https://coreyjames.ai/technologies/intel-oneapi/); [oneDNN](https://coreyjames.ai/technologies/onednn/); [CUTLASS](https://coreyjames.ai/technologies/cutlass/); [Bringing a serving stack up on new silicon](https://coreyjames.ai/stories/bringing-a-serving-stack-up-on-new-silicon/); [SYCL](https://coreyjames.ai/technologies/sycl/)
- NVIDIA Blackwell development: [NVIDIA Blackwell GPUs](https://coreyjames.ai/technologies/nvidia-blackwell-gpus/); [CUDA C++](https://coreyjames.ai/technologies/cuda-c-plus-plus/); [Built for Blackwell](https://coreyjames.ai/stories/built-for-blackwell/); [Inference Fabric](https://coreyjames.ai/systems/inference-fabric/); [Thirty-eight metrics in a dozen passes](https://coreyjames.ai/stories/thirty-eight-metrics-in-a-dozen-passes/)
- GPU profiling (Nsight): [NVIDIA Nsight Systems](https://coreyjames.ai/technologies/nvidia-nsight-systems/); [The thirteen percent that was never missing](https://coreyjames.ai/stories/the-thirteen-percent-that-was-never-missing/); [Inference Fabric](https://coreyjames.ai/systems/inference-fabric/); [Machinery I built and then did not need](https://coreyjames.ai/stories/machinery-i-built-and-then-did-not-need/)
- high-throughput batch inference: [Inference Fabric](https://coreyjames.ai/systems/inference-fabric/); [High-throughput batch inference and embeddings](https://coreyjames.ai/capabilities/high-throughput-batch-inference/); [Starting full and ending clean](https://coreyjames.ai/stories/starting-full-and-ending-clean/); [NVIDIA Blackwell GPUs](https://coreyjames.ai/technologies/nvidia-blackwell-gpus/); [The workload that shaped the platform](https://coreyjames.ai/stories/the-workload-that-shaped-the-platform/)
- token-aware scheduling: [A message broker that understands tokens](https://coreyjames.ai/stories/a-message-broker-that-understands-tokens/); [The token as the unit](https://coreyjames.ai/ideas/the-token-as-the-unit/); [Token-aware GPU batch scheduling](https://coreyjames.ai/technologies/token-aware-gpu-batch-scheduling/); [The workload that shaped the platform](https://coreyjames.ai/stories/the-workload-that-shaped-the-platform/); [High-throughput batch inference and embeddings](https://coreyjames.ai/capabilities/high-throughput-batch-inference/)
- message broker and queue design: [NVMe (cold-tier storage)](https://coreyjames.ai/technologies/nvme-cold-tier-storage/); [A message broker that understands tokens](https://coreyjames.ai/stories/a-message-broker-that-understands-tokens/); [Token-aware GPU batch scheduling](https://coreyjames.ai/technologies/token-aware-gpu-batch-scheduling/); [Disconnected producers and consumers](https://coreyjames.ai/ideas/disconnected-producers-and-consumers/); [Family Two: Durable work, tombstones and lossless recovery](https://coreyjames.ai/research/family-two-durable-work-and-lossless-recovery/)
- exactly-once delivery and crash recovery: [The workload that shaped the platform](https://coreyjames.ai/stories/the-workload-that-shaped-the-platform/); [ComfyUI](https://coreyjames.ai/technologies/comfyui/); [Business-provider APIs rebuilt as governed agent tools](https://coreyjames.ai/technologies/business-provider-apis-rebuilt-as-governed-agent-tools/); [Business systems as safe agent tools](https://coreyjames.ai/capabilities/business-systems-as-safe-agent-tools/); [Agents with real tools and no keys](https://coreyjames.ai/stories/agents-with-real-tools-and-no-keys/)
- append-only log design: [Append-only log and journal design](https://coreyjames.ai/technologies/append-only-log-and-journal-design/); [Append-only, with tombstones and lineage](https://coreyjames.ai/ideas/append-only-with-lineage/); [Storage that refuses to become a coordinator](https://coreyjames.ai/stories/storage-that-refuses-to-become-a-coordinator/); [Done means delivered](https://coreyjames.ai/stories/done-means-delivered/); [Inference Fabric](https://coreyjames.ai/systems/inference-fabric/)
- PostgreSQL performance tuning: [Database performance diagnosis](https://coreyjames.ai/technologies/database-performance-diagnosis/); [PostgreSQL bulk-update engineering](https://coreyjames.ai/technologies/postgresql-bulk-update-engineering/); [196 deadlocks to zero](https://coreyjames.ai/stories/196-deadlocks-to-zero/); [Twenty thousand rows a second](https://coreyjames.ai/stories/twenty-thousand-rows-a-second/); [PostgreSQL Fabric](https://coreyjames.ai/systems/postgresql-fabric/)
- database admission control: [Pricing writes by what they cost the database](https://coreyjames.ai/stories/pricing-writes-by-what-they-cost/); [Admission control and rate limiting](https://coreyjames.ai/technologies/admission-control-and-rate-limiting/); [PostgreSQL Fabric](https://coreyjames.ai/systems/postgresql-fabric/); [Policy-as-code enforcement](https://coreyjames.ai/technologies/policy-as-code-enforcement/); [Storage that refuses to become a coordinator](https://coreyjames.ai/stories/storage-that-refuses-to-become-a-coordinator/)
- vector search and ANN indexes: [FAISS](https://coreyjames.ai/technologies/faiss/); [GPU vector search and corpus analytics](https://coreyjames.ai/capabilities/gpu-vector-search-and-corpus-analytics/); [pgvector](https://coreyjames.ai/technologies/pgvector/); [Redistribution inside the database](https://coreyjames.ai/stories/redistribution-inside-the-database/); [Family Four: Self-verifying vector and GPU retrieval substrate](https://coreyjames.ai/research/family-four-self-verifying-vector-substrate/)
- embedding analytics and cluster metrics: [Thirty-eight metrics in a dozen passes](https://coreyjames.ai/stories/thirty-eight-metrics-in-a-dozen-passes/); [Where a cluster sits](https://coreyjames.ai/stories/where-a-cluster-sits/); [Physics as a lens on embeddings](https://coreyjames.ai/stories/physics-as-a-lens-on-embeddings/); [Gaussian-kernel cluster analysis](https://coreyjames.ai/technologies/gaussian-kernel-cluster-analysis/); [Removing the sampling caps](https://coreyjames.ai/stories/removing-the-sampling-caps/)
- agent orchestration: [AgentOS](https://coreyjames.ai/systems/agentos/); [Multi-agent system architecture and orchestration](https://coreyjames.ai/technologies/multi-agent-system-architecture-and-orchestration/); [Why long agent runs fail](https://coreyjames.ai/stories/why-long-agent-runs-fail/); [Model Context Protocol (MCP)](https://coreyjames.ai/technologies/model-context-protocol-mcp/); [From a conversation to a work order](https://coreyjames.ai/stories/from-a-conversation-to-a-work-order/)
- multi-agent governance: [Agent Client Protocol (ACP)](https://coreyjames.ai/technologies/agent-client-protocol-acp/); [Policy-as-code enforcement](https://coreyjames.ai/technologies/policy-as-code-enforcement/); [AgentOS](https://coreyjames.ai/systems/agentos/); [Family One: Deterministic external governance of autonomous AI work](https://coreyjames.ai/research/family-one-deterministic-external-governance/); [Separation of duties for AI work](https://coreyjames.ai/stories/separation-of-duties-for-ai-work/)
- evidence-gated acceptance: [Closed by evidence](https://coreyjames.ai/stories/closed-by-evidence/); [Family One: Deterministic external governance of autonomous AI work](https://coreyjames.ai/research/family-one-deterministic-external-governance/); [Governed AI-assisted software engineering](https://coreyjames.ai/technologies/governed-ai-assisted-software-engineering/); [Independent verification and adversarial review](https://coreyjames.ai/technologies/independent-verification-and-adversarial-review/); [Evidence-gated API integration](https://coreyjames.ai/technologies/evidence-gated-api-integration/)
- separation of duties for AI work: [AI subscriptions as governed workers](https://coreyjames.ai/capabilities/use-the-ai-subscriptions-you-already-have/); [Multi-agent system architecture and orchestration](https://coreyjames.ai/technologies/multi-agent-system-architecture-and-orchestration/); [Separation of duties for AI work](https://coreyjames.ai/stories/separation-of-duties-for-ai-work/); [Role contracts, with vendors as data](https://coreyjames.ai/ideas/role-contracts-not-vendors/); [Agent tool-execution safety and sandboxing](https://coreyjames.ai/technologies/agent-tool-execution-safety-and-sandboxing/)
- sandboxing and container hardening: [Docker](https://coreyjames.ai/technologies/docker/); [A coding agent that never holds the keys](https://coreyjames.ai/stories/a-coding-agent-that-never-holds-the-keys/); [Agent tool-execution safety and sandboxing](https://coreyjames.ai/technologies/agent-tool-execution-safety-and-sandboxing/); [Linux security primitives](https://coreyjames.ai/technologies/linux-security-primitives/); [Agent Swarm](https://coreyjames.ai/systems/agent-swarm/)
- MCP server design: [TypeScript](https://coreyjames.ai/technologies/typescript/); [rmcp (Rust MCP SDK)](https://coreyjames.ai/technologies/rmcp-rust-mcp-sdk/); [Model Context Protocol (MCP)](https://coreyjames.ai/technologies/model-context-protocol-mcp/); [JSON-RPC](https://coreyjames.ai/technologies/json-rpc/); [AI teammates inside Microsoft Teams and team chat](https://coreyjames.ai/capabilities/ai-teammates-in-teams-and-chat/)
- tool synthesis and governed API integration: [Tools that build tools, under governance](https://coreyjames.ai/stories/tools-that-build-tools-under-governance/); [Evidence-gated API integration](https://coreyjames.ai/technologies/evidence-gated-api-integration/); [Business systems as safe agent tools](https://coreyjames.ai/capabilities/business-systems-as-safe-agent-tools/); [Tools are the platform](https://coreyjames.ai/ideas/tools-are-the-platform/); [Business-provider APIs rebuilt as governed agent tools](https://coreyjames.ai/technologies/business-provider-apis-rebuilt-as-governed-agent-tools/)
- OAuth 2.0 and identity architecture: [OAuth 2.0](https://coreyjames.ai/technologies/oauth-2-0/); [FastAPI](https://coreyjames.ai/technologies/fastapi/); [Auth and access plane](https://coreyjames.ai/systems/auth-and-access-plane/); [Python](https://coreyjames.ai/technologies/python/); [Identity and access management for AI agents](https://coreyjames.ai/technologies/identity-and-access-management-for-ai-agents/)
- zero-trust and zero-inbound networking: [Zero-inbound network exposure](https://coreyjames.ai/technologies/zero-inbound-network-exposure/); [Site reliability engineering](https://coreyjames.ai/technologies/site-reliability-engineering/); [Cloudflare Tunnel](https://coreyjames.ai/technologies/cloudflare-tunnel/); [Federated compute provisioning](https://coreyjames.ai/technologies/federated-compute-provisioning/); [Auth and access plane](https://coreyjames.ai/systems/auth-and-access-plane/)
- Microsoft 365, Graph and Entra integration: [Solving the digital-employee identity](https://coreyjames.ai/stories/solving-the-digital-employee-identity/); [Identity and access management for AI agents](https://coreyjames.ai/technologies/identity-and-access-management-for-ai-agents/); [Microsoft Entra ID](https://coreyjames.ai/technologies/microsoft-entra-id/); [Bash](https://coreyjames.ai/technologies/bash/); [Microsoft Graph](https://coreyjames.ai/technologies/microsoft-graph/)
- Microsoft Teams real-time media: [A self-hosted colleague in a Teams call](https://coreyjames.ai/stories/a-self-hosted-colleague-in-a-teams-call/); [AI teammates inside Microsoft Teams and team chat](https://coreyjames.ai/capabilities/ai-teammates-in-teams-and-chat/); [Vendor-integration reverse engineering](https://coreyjames.ai/technologies/vendor-integration-reverse-engineering/); [One percent network, seventy-eight percent GPU](https://coreyjames.ai/stories/one-percent-network-seventy-eight-percent-gpu/); [C#](https://coreyjames.ai/technologies/c-sharp/)
- real-time voice AI: [Conversational turn-taking and barge-in](https://coreyjames.ai/technologies/conversational-turn-taking-and-barge-in/); [AI teammates inside Microsoft Teams and team chat](https://coreyjames.ai/capabilities/ai-teammates-in-teams-and-chat/); [Voice agents in calls and meetings](https://coreyjames.ai/capabilities/voice-agents-in-meetings/); [Real-time audio pipeline engineering](https://coreyjames.ai/technologies/real-time-audio-pipeline-engineering/); [The same brain, typed or spoken](https://coreyjames.ai/stories/the-same-brain-typed-or-spoken/)
- speech recognition and synthesis pipelines: [Voice agents in calls and meetings](https://coreyjames.ai/capabilities/voice-agents-in-meetings/); [Helix](https://coreyjames.ai/systems/helix/); [ONNX Runtime](https://coreyjames.ai/technologies/onnx-runtime/); [Speech recognition model](https://coreyjames.ai/technologies/speech-recognition-model/); [Speech synthesis model](https://coreyjames.ai/technologies/speech-synthesis-model/)
- Rust systems engineering: [Tokio](https://coreyjames.ai/technologies/tokio/); [Rust](https://coreyjames.ai/technologies/rust/); [Closed by evidence](https://coreyjames.ai/stories/closed-by-evidence/); [PostgreSQL Fabric](https://coreyjames.ai/systems/postgresql-fabric/); [Built for Blackwell](https://coreyjames.ai/stories/built-for-blackwell/)
- C++ and native ports with parity proof: [Large-scale refactoring with parity proof](https://coreyjames.ai/technologies/large-scale-refactoring-with-parity-proof/); [pybind11](https://coreyjames.ai/technologies/pybind11/); [C++](https://coreyjames.ai/technologies/c-plus-plus/); [Directing AI coding agents at program scale](https://coreyjames.ai/stories/directing-ai-coding-agents-at-program-scale/); [Governed AI-assisted software engineering](https://coreyjames.ai/technologies/governed-ai-assisted-software-engineering/)
- distributed systems correctness: [A self-hosted colleague in a Teams call](https://coreyjames.ai/stories/a-self-hosted-colleague-in-a-teams-call/); [Reconcile, never resubmit](https://coreyjames.ai/ideas/reconcile-never-resubmit/); [Millions of queued items that were not there](https://coreyjames.ai/stories/millions-of-queued-items-that-were-not-there/); [Lua](https://coreyjames.ai/technologies/lua/); [Cache synchronisation and leader election](https://coreyjames.ai/technologies/cache-synchronisation-and-leader-election/)
- observability and alerting: [Site reliability engineering](https://coreyjames.ai/technologies/site-reliability-engineering/); [Observability and alerting](https://coreyjames.ai/technologies/observability-and-alerting/); [Prometheus](https://coreyjames.ai/technologies/prometheus/); [Grafana](https://coreyjames.ai/technologies/grafana/); [Shapes have a lifecycle](https://coreyjames.ai/stories/shapes-have-a-lifecycle/)
- cost attribution for AI work: [Cost attribution for AI work](https://coreyjames.ai/technologies/cost-attribution-for-ai-work/); [Observability and alerting](https://coreyjames.ai/technologies/observability-and-alerting/); [What the work cost, to the penny](https://coreyjames.ai/stories/what-the-work-cost/); [AI subscriptions as governed workers](https://coreyjames.ai/capabilities/use-the-ai-subscriptions-you-already-have/); [AgentOS](https://coreyjames.ai/systems/agentos/)
- AI-assisted engineering methodology: [Governed AI-assisted software engineering](https://coreyjames.ai/technologies/governed-ai-assisted-software-engineering/); [Closed by evidence](https://coreyjames.ai/stories/closed-by-evidence/); [Governed autonomous engineering](https://coreyjames.ai/capabilities/governed-autonomous-engineering/); [Directing AI coding agents at program scale](https://coreyjames.ai/stories/directing-ai-coding-agents-at-program-scale/); [Two thousand concurrent workers](https://coreyjames.ai/stories/two-thousand-concurrent-workers/)
- technical leadership of agent teams: [Governed AI-assisted software engineering](https://coreyjames.ai/technologies/governed-ai-assisted-software-engineering/); [Closed by evidence](https://coreyjames.ai/stories/closed-by-evidence/); [Multi-agent system architecture and orchestration](https://coreyjames.ai/technologies/multi-agent-system-architecture-and-orchestration/); [Rewriting an engine extension without changing a token](https://coreyjames.ai/stories/rewriting-an-engine-extension-without-changing-a-token/); [Directing AI coding agents at program scale](https://coreyjames.ai/stories/directing-ai-coding-agents-at-program-scale/)

## Technologies (177), each with the systems and stories where it matters: https://coreyjames.ai/technologies/
- datastore (6): [FAISS](https://coreyjames.ai/technologies/faiss/); [Garnet](https://coreyjames.ai/technologies/garnet/); [pgvector](https://coreyjames.ai/technologies/pgvector/); [PostgreSQL](https://coreyjames.ai/technologies/postgresql/); [Redis](https://coreyjames.ai/technologies/redis/); [SQLite](https://coreyjames.ai/technologies/sqlite/)
- discipline (82): [Admission control and rate limiting](https://coreyjames.ai/technologies/admission-control-and-rate-limiting/); [Agent tool qualification testing](https://coreyjames.ai/technologies/agent-tool-qualification-testing/); [Agent tool-execution safety and sandboxing](https://coreyjames.ai/technologies/agent-tool-execution-safety-and-sandboxing/); [Append-only log and journal design](https://coreyjames.ai/technologies/append-only-log-and-journal-design/); [Application security hardening and threat modelling](https://coreyjames.ai/technologies/application-security-hardening-and-threat-modelling/); [Applied cryptography](https://coreyjames.ai/technologies/applied-cryptography/); [Approximate nearest-neighbour search](https://coreyjames.ai/technologies/approximate-nearest-neighbour-search/); [Binary wire protocol design](https://coreyjames.ai/technologies/binary-wire-protocol-design/); [Budget enforcement and cost governance](https://coreyjames.ai/technologies/budget-enforcement-and-cost-governance/); [Business-provider APIs rebuilt as governed agent tools](https://coreyjames.ai/technologies/business-provider-apis-rebuilt-as-governed-agent-tools/); [Cache synchronisation and leader election](https://coreyjames.ai/technologies/cache-synchronisation-and-leader-election/); [Capacity engineering, backpressure and load testing](https://coreyjames.ai/technologies/capacity-engineering-backpressure-and-load-testing/); [Catalog versioning and drift detection](https://coreyjames.ai/technologies/catalog-versioning-and-drift-detection/); [Cluster-health metrics on the GPU](https://coreyjames.ai/technologies/cluster-health-metrics-on-the-gpu/); [Code generation and metaprogramming](https://coreyjames.ai/technologies/code-generation-and-metaprogramming/); [Code graph and static dependency analysis](https://coreyjames.ai/technologies/code-graph-and-static-dependency-analysis/); [Context engineering and persistent agent memory](https://coreyjames.ai/technologies/context-engineering-and-persistent-agent-memory/); [Contract-first interface design](https://coreyjames.ai/technologies/contract-first-interface-design/); [Conversational turn-taking and barge-in](https://coreyjames.ai/technologies/conversational-turn-taking-and-barge-in/); [Cost attribution for AI work](https://coreyjames.ai/technologies/cost-attribution-for-ai-work/); [CPU reference implementations and GPU equivalence testing](https://coreyjames.ai/technologies/cpu-reference-implementations-and-gpu-equivalence-testing/); [Credential brokering and secrets management](https://coreyjames.ai/technologies/credential-brokering-and-secrets-management/); [Database performance diagnosis](https://coreyjames.ai/technologies/database-performance-diagnosis/); [Declarative tool and agent definition](https://coreyjames.ai/technologies/declarative-tool-and-agent-definition/); [Determinism and numerical reproducibility](https://coreyjames.ai/technologies/determinism-and-numerical-reproducibility/); [Distributed job orchestration](https://coreyjames.ai/technologies/distributed-job-orchestration/); [Distributed-systems reliability](https://coreyjames.ai/technologies/distributed-systems-reliability/); [Evidence-gated acceptance](https://coreyjames.ai/technologies/evidence-gated-acceptance/); [Evidence-gated API integration](https://coreyjames.ai/technologies/evidence-gated-api-integration/); [Evidence-grounded LLM output](https://coreyjames.ai/technologies/evidence-grounded-llm-output/); [Federated compute provisioning](https://coreyjames.ai/technologies/federated-compute-provisioning/); [Gaussian-kernel cluster analysis](https://coreyjames.ai/technologies/gaussian-kernel-cluster-analysis/); [Governed AI-assisted software engineering](https://coreyjames.ai/technologies/governed-ai-assisted-software-engineering/); [GPU admission control and co-residency](https://coreyjames.ai/technologies/gpu-admission-control-and-co-residency/); [GPU inference engine engineering](https://coreyjames.ai/technologies/gpu-inference-engine-engineering/); [GPU memory management](https://coreyjames.ai/technologies/gpu-memory-management/); [GPU numerical kernel design](https://coreyjames.ai/technologies/gpu-numerical-kernel-design/); [GPU profiling and tracing](https://coreyjames.ai/technologies/gpu-profiling-and-tracing/); [Grammar-constrained decoding](https://coreyjames.ai/technologies/grammar-constrained-decoding/); [Hidden-state extraction](https://coreyjames.ai/technologies/hidden-state-extraction/); [Hybrid retrieval](https://coreyjames.ai/technologies/hybrid-retrieval/); [Idempotency, exactly-once effects and crash recovery](https://coreyjames.ai/technologies/idempotency-exactly-once-effects-and-crash-recovery/); [Identity and access management for AI agents](https://coreyjames.ai/technologies/identity-and-access-management-for-ai-agents/); [Independent verification and adversarial review](https://coreyjames.ai/technologies/independent-verification-and-adversarial-review/); [KV-cache engineering](https://coreyjames.ai/technologies/kv-cache-engineering/); [Large-scale GPU data pipelines](https://coreyjames.ai/technologies/large-scale-gpu-data-pipelines/); [Large-scale refactoring with parity proof](https://coreyjames.ai/technologies/large-scale-refactoring-with-parity-proof/); [LD_PRELOAD symbol interposition](https://coreyjames.ai/technologies/ld-preload-symbol-interposition/); [Lease-based work claiming](https://coreyjames.ai/technologies/lease-based-work-claiming/); [Link-time plugin registration](https://coreyjames.ai/technologies/link-time-plugin-registration/); [Linux security primitives](https://coreyjames.ai/technologies/linux-security-primitives/); [LLM fine-tuning](https://coreyjames.ai/technologies/llm-fine-tuning/); [MCP tool design](https://coreyjames.ai/technologies/mcp-tool-design/); [ML evaluation methodology](https://coreyjames.ai/technologies/ml-evaluation-methodology/); [Model quantization](https://coreyjames.ai/technologies/model-quantization/); [Multi-agent system architecture and orchestration](https://coreyjames.ai/technologies/multi-agent-system-architecture-and-orchestration/); [Multi-vendor LLM integration](https://coreyjames.ai/technologies/multi-vendor-llm-integration/); [Network egress confinement](https://coreyjames.ai/technologies/network-egress-confinement/); [Observability and alerting](https://coreyjames.ai/technologies/observability-and-alerting/); [Performance engineering and benchmarking](https://coreyjames.ai/technologies/performance-engineering-and-benchmarking/); [Policy-as-code enforcement](https://coreyjames.ai/technologies/policy-as-code-enforcement/); [PostgreSQL bulk-update engineering](https://coreyjames.ai/technologies/postgresql-bulk-update-engineering/); [Prompt-injection-aware tooling](https://coreyjames.ai/technologies/prompt-injection-aware-tooling/); [Provider-terms compliance in model routing](https://coreyjames.ai/technologies/provider-terms-compliance-in-model-routing/); [Real-time audio pipeline engineering](https://coreyjames.ai/technologies/real-time-audio-pipeline-engineering/); [Research data forensics](https://coreyjames.ai/technologies/research-data-forensics/); [Reversible infrastructure mutation](https://coreyjames.ai/technologies/reversible-infrastructure-mutation/); [Schema and migration discipline](https://coreyjames.ai/technologies/schema-and-migration-discipline/); [Schema-agnostic data access and SQL-injection defence](https://coreyjames.ai/technologies/schema-agnostic-data-access-and-sql-injection-defence/); [Schema-driven build-time code generation](https://coreyjames.ai/technologies/schema-driven-build-time-code-generation/); [Site reliability engineering](https://coreyjames.ai/technologies/site-reliability-engineering/); [State-machine design for multi-agent workflows](https://coreyjames.ai/technologies/state-machine-design-for-multi-agent-workflows/); [Static analysis as tests](https://coreyjames.ai/technologies/static-analysis-as-tests/); [Streaming statistics in CUDA](https://coreyjames.ai/technologies/streaming-statistics-in-cuda/); [Synthetic training-data generation](https://coreyjames.ai/technologies/synthetic-training-data-generation/); [Tiled pairwise kernels](https://coreyjames.ai/technologies/tiled-pairwise-kernels/); [Token-aware GPU batch scheduling](https://coreyjames.ai/technologies/token-aware-gpu-batch-scheduling/); [Vendor-integration reverse engineering](https://coreyjames.ai/technologies/vendor-integration-reverse-engineering/); [Web crawling and document extraction](https://coreyjames.ai/technologies/web-crawling-and-document-extraction/); [Webhook ingestion and verification](https://coreyjames.ai/technologies/webhook-ingestion-and-verification/); [Write-behind persistence](https://coreyjames.ai/technologies/write-behind-persistence/); [Zero-inbound network exposure](https://coreyjames.ai/technologies/zero-inbound-network-exposure/)
- framework (15): [Axum](https://coreyjames.ai/technologies/axum/); [CUTLASS](https://coreyjames.ai/technologies/cutlass/); [FastAPI](https://coreyjames.ai/technologies/fastapi/); [Handlebars](https://coreyjames.ai/technologies/handlebars/); [Hugging Face Transformers](https://coreyjames.ai/technologies/hugging-face-transformers/); [Microsoft real-time media SDK](https://coreyjames.ai/technologies/microsoft-real-time-media-sdk/); [pybind11](https://coreyjames.ai/technologies/pybind11/); [PyO3](https://coreyjames.ai/technologies/pyo3/); [PyTorch](https://coreyjames.ai/technologies/pytorch/); [ratatui](https://coreyjames.ai/technologies/ratatui/); [React](https://coreyjames.ai/technologies/react/); [rmcp (Rust MCP SDK)](https://coreyjames.ai/technologies/rmcp-rust-mcp-sdk/); [sqlx](https://coreyjames.ai/technologies/sqlx/); [Tokio](https://coreyjames.ai/technologies/tokio/); [tree-sitter](https://coreyjames.ai/technologies/tree-sitter/)
- hardware (3): [Intel Arc GPUs](https://coreyjames.ai/technologies/intel-arc-gpus/); [NVIDIA Blackwell GPUs](https://coreyjames.ai/technologies/nvidia-blackwell-gpus/); [NVMe (cold-tier storage)](https://coreyjames.ai/technologies/nvme-cold-tier-storage/)
- language (11): [Bash](https://coreyjames.ai/technologies/bash/); [C#](https://coreyjames.ai/technologies/c-sharp/); [C++](https://coreyjames.ai/technologies/c-plus-plus/); [CUDA C++](https://coreyjames.ai/technologies/cuda-c-plus-plus/); [JavaScript](https://coreyjames.ai/technologies/javascript/); [Lua](https://coreyjames.ai/technologies/lua/); [Python](https://coreyjames.ai/technologies/python/); [Rust](https://coreyjames.ai/technologies/rust/); [SQL](https://coreyjames.ai/technologies/sql/); [TypeScript](https://coreyjames.ai/technologies/typescript/); [YAML](https://coreyjames.ai/technologies/yaml/)
- model (6): [Model families and types](https://coreyjames.ai/technologies/model-families-and-types/); [Multilingual embedding model](https://coreyjames.ai/technologies/multilingual-embedding-model/); [Reranker model](https://coreyjames.ai/technologies/reranker-model/); [Speech recognition model](https://coreyjames.ai/technologies/speech-recognition-model/); [Speech synthesis model](https://coreyjames.ai/technologies/speech-synthesis-model/); [Text embedding models](https://coreyjames.ai/technologies/text-embedding-models/)
- platform (28): [Cargo](https://coreyjames.ai/technologies/cargo/); [Chrome extensions (Manifest V3)](https://coreyjames.ai/technologies/chrome-extensions-manifest-v3/); [Cloudflare Tunnel](https://coreyjames.ai/technologies/cloudflare-tunnel/); [Cloudflare Workers](https://coreyjames.ai/technologies/cloudflare-workers/); [ComfyUI](https://coreyjames.ai/technologies/comfyui/); [Docker](https://coreyjames.ai/technologies/docker/); [FFmpeg](https://coreyjames.ai/technologies/ffmpeg/); [Git](https://coreyjames.ai/technologies/git/); [GitLab](https://coreyjames.ai/technologies/gitlab/); [Grafana](https://coreyjames.ai/technologies/grafana/); [Hugging Face Hub](https://coreyjames.ai/technologies/hugging-face-hub/); [Mattermost](https://coreyjames.ai/technologies/mattermost/); [Meta Graph API](https://coreyjames.ai/technologies/meta-graph-api/); [Microsoft Agent 365](https://coreyjames.ai/technologies/microsoft-agent-365/); [Microsoft Azure](https://coreyjames.ai/technologies/microsoft-azure/); [Microsoft Copilot Studio](https://coreyjames.ai/technologies/microsoft-copilot-studio/); [Microsoft Entra ID](https://coreyjames.ai/technologies/microsoft-entra-id/); [Microsoft Graph](https://coreyjames.ai/technologies/microsoft-graph/); [Microsoft Teams](https://coreyjames.ai/technologies/microsoft-teams/); [NGINX](https://coreyjames.ai/technologies/nginx/); [NVIDIA Nsight Systems](https://coreyjames.ai/technologies/nvidia-nsight-systems/); [Prometheus](https://coreyjames.ai/technologies/prometheus/); [Public environmental data APIs](https://coreyjames.ai/technologies/public-environmental-data-apis/); [Reddit](https://coreyjames.ai/technologies/reddit/); [RunPod](https://coreyjames.ai/technologies/runpod/); [systemd](https://coreyjames.ai/technologies/systemd/); [tmux](https://coreyjames.ai/technologies/tmux/); [X API](https://coreyjames.ai/technologies/x-api/)
- protocol (17): [ACME](https://coreyjames.ai/technologies/acme/); [Agent Client Protocol (ACP)](https://coreyjames.ai/technologies/agent-client-protocol-acp/); [CBOR](https://coreyjames.ai/technologies/cbor/); [GraphQL](https://coreyjames.ai/technologies/graphql/); [H.264](https://coreyjames.ai/technologies/h-264/); [JSON Schema](https://coreyjames.ai/technologies/json-schema/); [JSON Web Token (JWT)](https://coreyjames.ai/technologies/json-web-token-jwt/); [JSON-RPC](https://coreyjames.ai/technologies/json-rpc/); [MessagePack](https://coreyjames.ai/technologies/messagepack/); [Model Context Protocol (MCP)](https://coreyjames.ai/technologies/model-context-protocol-mcp/); [OAuth 2.0](https://coreyjames.ai/technologies/oauth-2-0/); [OpenAI-compatible API](https://coreyjames.ai/technologies/openai-compatible-api/); [OpenAPI](https://coreyjames.ai/technologies/openapi/); [TLS and mTLS](https://coreyjames.ai/technologies/tls-and-mtls/); [Unix domain sockets](https://coreyjames.ai/technologies/unix-domain-sockets/); [WebAuthn](https://coreyjames.ai/technologies/webauthn/); [WebSocket](https://coreyjames.ai/technologies/websocket/)
- runtime (9): [CUDA Driver API](https://coreyjames.ai/technologies/cuda-driver-api/); [Hugging Face Text Embeddings Inference](https://coreyjames.ai/technologies/hugging-face-text-embeddings-inference/); [Intel oneAPI](https://coreyjames.ai/technologies/intel-oneapi/); [NVIDIA TensorRT](https://coreyjames.ai/technologies/nvidia-tensorrt/); [NVIDIA TensorRT-LLM](https://coreyjames.ai/technologies/nvidia-tensorrt-llm/); [oneDNN](https://coreyjames.ai/technologies/onednn/); [ONNX Runtime](https://coreyjames.ai/technologies/onnx-runtime/); [SYCL](https://coreyjames.ai/technologies/sycl/); [vLLM](https://coreyjames.ai/technologies/vllm/)