Selected work
The RAG systems are public, with reproducible evals; the cluster and agent platform are private infrastructure. Each entry leads with the outcome, and every project metric is labeled measured or estimated.
Cited, revision-aware retrieval over 321,124 real IETF RFC chunks, with zero LLM in the answer path. It resolves the real RFC 2616 obsoletion exactly, six ways into RFC 7230 through 7235, by walking a supersession graph.
Routes a question across subject-specialized knowledge modules, binds every claim to a source span, and abstains instead of guessing. Measured head-to-head against a naive RAG baseline on the identical corpus.
A rule-based checker that flags made-up claims in AI-generated text. Tested against a public benchmark, it caught fewer than 3 in 100 fabrications. I put that number at the top of the README instead of burying it, because the failure maps exactly where this whole class of tool stops working.
Twelve small, offline, deterministic tools (one per RAG pipeline stage) that turn silent retrieval failures into loud, named signals. Each ships a red/green self-test that reproduces a real failure, catches it, and clears a clean case so the catch isn't rigged.
Turns a folder of documents into a searchable Qdrant index in one line: GPU-optional embeddings, optional GPU OCR for scanned PDFs, and fail-closed device selection. Installable from source.
A queryable graph of which RFCs obsolete which, which errata correct which, and which entities co-occur. Built over the real entity-mention stream from the RFC-corpus RAG.
Detects thin evidence in an answer, ranks where to hunt for more, retries once inside a bounded cycle, and audits every attempt. Proven against the real 321k-chunk RFC corpus.
A second, independent vertical on the same architecture: cited answers over Bitcoin, Ethereum, Solana, Monero, Polygon, and Cardano protocol docs, plus a router-calibration finding the reliability gates exist to catch.
A modular, offline-capable multi-agent platform (OpenAI-compatible gateway, control plane, autonomy engine, RAG, and IDE integration) with an automated adversarial peer-review step and zero hosted-LLM dependencies.
A self-built two-node multi-GPU cluster serving local models with a privacy-first architecture, thermal-aware autonomy, and GPU-wedge diagnosis and recovery. The RTX 3090 node is down with a hardware fault until I repair it, so the RTX 5090 serves alone for now.