I'm an AI/ML Engineer specializing in production agentic AI systems, LLM evaluation infrastructure, and RAG pipelines — with 3+ years of experience shipping AI systems across FinTech, GovTech, and AI Research.
- 🎓 MS Computer Science (Machine Learning) @ George Mason University — GPA 3.7 (May 2026)
- 🤖 Research Assistant building an LLM-powered multi-agent code analysis platform @ GMU AI Code Analysis Lab
- 🚀 Previously AI Engineer @ InterAgency Inc. — built a production RAG-powered permitting chatbot for DC government workflows
- 🔍 Expert in LLM-as-judge evaluation, hallucination detection, benchmark construction, and agent harness design
- 🛠️ Active open-source contributor to LiteLLM (merged PR #31574) — unified LLM gateway used by NASA, Adobe, Rocket Money
- 🍎 Fine-tuned Qwen2.5-Coder-7B with LoRA using Apple's MLX framework on Apple Silicon
- 📜 NVIDIA Certified Associate — Generative AI and LLMs (2025)
- 📍 Based in Fairfax, VA — open to relocation
- 💼 Currently open to full-time AI/ML Engineering roles
| Area | Tools |
|---|---|
| Agentic AI Systems | LangChain, LangGraph, CrewAI, MCP Servers, Tool Use, Agent Harness Design |
| LLM Evaluation | LLM-as-Judge, Hallucination Detection, MLflow, Benchmark Construction, Failure Mode Analysis |
| RAG Pipelines | LlamaIndex, ChromaDB, Pinecone, Semantic Chunking, Vector Search |
| Production AI | FastAPI, Flask, AWS ECS Fargate, Docker, Kubernetes |
| ML Frameworks | PyTorch, TensorFlow, Scikit-learn, Hugging Face, MLX |
| Full Stack | React.js, Node.js, TypeScript, PostgreSQL, Redis |
Perry — DC Permit RAG Chatbot Production RAG-powered permitting chatbot serving 88+ DC permit documents at sub-3s latency. Built with Claude API, LlamaIndex, ChromaDB, Flask, and AWS.
AST-Grep Agentic Code Analysis Platform Multi-agent code analysis system converting plain-English design rules into executable AST patterns across 20+ languages. Fine-tuned Qwen2.5-Coder-7B via LoRA on MLX, achieving 86% syntax validity through an automated verify-and-retry loop.
NLBot — Agentic Voice Robot Planner Agentic human-robot system where a Claude Sonnet planning agent decomposes voice commands into structured JSON task plans with human-in-the-loop approval and hallucination mitigation.
Architectural Document Analysis with LLMs Multimodal agentic system combining LLaVA 7B and Llama 3.2 for architectural drawing analysis. Achieved F1=0.82 and SUS usability score of 74.3.
ConnectEd — AI Student Mentor Platform Agentic mentor matching system using Groq Llama 3.1 70B achieving 100% successful pairing rate through semantic similarity scoring.
LiteLLM — Open Source Contribution Merged PR #31574 to LiteLLM, the unified LLM gateway (34K+ GitHub stars) used by NASA, Adobe, Rocket Money, and Samsara.
Multi-Class Obesity Risk Prediction ML classification system comparing XGBoost (98.5%), Random Forest (98.4%), Decision Tree (96.9%), SVM (94.6%) with SHAP explainability analysis.
Languages
AI/ML
Cloud & Infrastructure
