AI AGENT SYSTEMS
Building ReAct agents with tool use, sub-agent delegation, and persistent memory. From prototype to production.
Building production AI agents, RAG pipelines, and LLM evaluation systems.
Looking for entry-level AI Engineer roles — available May 2026
LLM OPS & EVALS
Shipping RAG systems, evaluation pipelines, and model-tier routing with measurable quality benchmarks.
Featured Projects
All Blog Posts
VitalSync: Health Intelligence Agent
Production ReAct agent on Gemini 2.5 Flash with 4 service-routed tools, LLM-as-judge regression suite, and async TTL-scoped memory with per-user observability.
1 technical deep-dive
RAG-on-Rails: Controlled-Experiment RAG Benchmark
From-scratch RAG benchmark with a single-variable harness across 30 experiments on MultiHop-RAG, lifting complete-evidence retrieval (AllGold@10) from 18.6% to 49.2% via hybrid search, reranking, and an agentic retrieve-judge-refine loop.
7 technical deep-dives
MiniClaw: Self-Hosted AI Agent
Open-ended ReAct agent with sub-agent delegation, planner-reducer split, map-reduce over 200+ Gmail messages, and hybrid SQLite FTS5 + sqlite-vec memory index.
Spidey Social: Serverless Real-Time Social App
1st place AWS category at Hack_NCState 2026 — serverless social app with geohash-indexed partitions for proximity queries and WebSockets for real-time messaging.
AI Research Analytics Platform
COI detection via co-authorship graph traversal across 23K+ relationships; hybrid rule-based + OpenAI publication classification over 800 papers; serving 35+ faculty at NC State.
FFA Course Delivery Platform
Full-stack multimedia course platform with enrollment, progress tracking, quiz delivery, and Firebase Auth + 3-tier RBAC — serving 500+ users for a 501(c)(3) nonprofit.