Work
Four public repos in the order I'd read them, then paid client work and earlier projects. Each number comes from the repo's README Results table, which links to its source file, and each README keeps a Methodology & limits section.
01 · Ship it · Measure it
Turns PDFs and scans into validated, searchable records, with an extraction eval that runs as a CI check.
95.5% weighted field-level score on a 28-fixture offline replay; the CI check fails below a 0.85 floor.
Try it locally, no API key
python scripts/eval_offline_replay.py --floor 0.85
02 · Secure it
Pre-registered research on indirect prompt injection through agent tool output, turned into a CI-ready auditor.
With a content-trust prompt scaffold in place, one payload variant still exfiltrated a canary in 10 of 10 runs on Groq-hosted Llama-3.3-70B (one provider, n=10). The recommended fix is in tool dispatch rather than the prompt; it has not been measured at 70B.
Try it locally, no API key
make repro
03 · Secure it
A library of opt-in decorators that wrap MCP tool logic with auth, a read-only SQL allowlist, caching, rate limits and tracing.
JWT/JWKS auth, a sqlglot read-only SQL allowlist and OpenTelemetry spans, with 627 collected tests.
Try it locally, no API key
pip install -e ".[dev]" && python examples/verified_cache.py
04 · Measure it
A 10-minute runnable code sample: eval gates, approval-bound writes and idempotent retries.
36 offline tests, no API key. A mutated label set fails the eval gate, and a duplicate retry is suppressed.
Try it locally, no API key
uv sync --group dev && uv run pytest
SMS lead qualification inside GoHighLevel, with Spanish detection and a bilingual human handoff. Client-reported 500+ inbound leads.
Paid AI application work for a confidential client. The repository is private; an authorized walkthrough is available on request.
Request a walkthrough →YAML-driven workflow automation backend with async job processing and SSE streaming.
Drop-in AI chat widget with a single script tag: RAG over pgvector, WebSocket streaming and Shadow DOM isolation.
Knowledge-graph retrieval for multi-hop questions, with a no-API-key BM25 mode and a side-by-side comparison against basic RAG.