Event-driven chat client for AI agent sessions over WebSocket. Handles event reordering, deduplication, reconnects, and replay through a strict pipeline from transport to render-only React UI.

Rust implementation of Git object primitives, modeling blobs, trees, commits, and tags with Git-compatible serialization and SHA-1 object IDs.

Parallel specialist agents review pull-request diffs, while a critic independently verifies each finding against the changed lines before one consolidated PR comment is produced.

Architected a distributed model inference pipeline across multiple AWS EC2 instances with private subnet networking, RPC worker coordination, and Terraform-automated provisioning. Documented design trade-offs and debugged latency under t3.micro constraints.

Multi-stage agent pipeline that converts scanned patient records into structured discharge summaries. OCR fallback (Tesseract) + PyMuPDF extraction, deterministic reconciliation for medication conflicts, and full provenance (document, page, source text) for clinical auditability.

Production RAG API for multi-format ingestion (PDFs, images, URLs, code). RAPTOR hierarchical indexing (UMAP + HDBSCAN), hybrid retrieval (pgvector HNSW + BM25 via RRF), Jina reranker, Gemini 2.0 Flash answers with citations, plus Prometheus/Grafana observability.

TypeScript-first note-taking app with rich text editing, organization, and search. Built for type safety, modularity, and performance — deployed on Vercel with Supabase and Google Gemini integration.

Intelligent resume analysis across 8+ ATS criteria (keywords, formatting, section completeness). Extracts structured data, returns 5+ actionable feedback points, and a 0–100 score that simulates recruiter filtering.