AI Engineer and M.S. Artificial Intelligence candidate at the University at Buffalo. I build production LLM systems across post-training, RAG, agents, evals, model serving, observability, and MLOps.
My work sits between applied ML, LLM engineering, and production systems. I fine-tune models, build retrieval and agent pipelines, design evaluation sets, instrument failures, and ship the serving and observability layers around them. The recurring theme is simple: an AI feature should be measurable, auditable, and useful outside a notebook.
Industry work across enterprise GenAI, RAG, agent workflows, APIs, data pipelines, evaluation, deployment, and production debugging.
Building enterprise Generative AI services grounded in client data, taking LLM features from prototype through evaluation, REST API integration, testing, production debugging, and deployment in a peer-reviewed codebase.
Shipped Python/LangChain/LangGraph RAG and multi-agent workflows for enterprise clients, cutting manual knowledge-retrieval effort by ~30%; built Python/SQL ETL and validation pipelines across 10K+ records.
Comparing a Regan-inspired statistical human-move model, Qwen3-8B, and Maia-3 on identical chess positions using Stockfish MultiPV, legal-move likelihood scoring, and controlled engine substitutions.
Systematizing failure modes of LLM/VLM-driven Physical AI across prompt/tool attacks, adversarial perception, leakage, and unsafe physical-world actions.
Four projects deliberately chosen to show different layers of AI engineering: post-training and retrieval, forward-deployed decision systems, agent orchestration, and production MLOps.
Fine-tuned Qwen3-4B with QLoRA/SFT/DPO and built hybrid BGE+BM25 retrieval with RRF/cross-encoder reranking and grounded citations. Releases are gated by automated evals rather than intuition.
A cold-chain decision-intelligence system that fuses minute-level telemetry, orders, and shipping documents, predicts cargo breaches early, reconciles conflicting records, and keeps every consequential action behind a human approval gate.
A LangGraph workflow that takes a problem from discovery through design and build, integrates MCP tools, checkpoints state, requires human approval before action, and traces cost and behavior across the agent graph.
An automated forecasting pipeline combining Elo ratings, gradient boosting, and Monte Carlo simulation, with Airflow orchestration, MLflow experiment tracking, Dockerized deployment, tracing, and drift monitoring.
I’m looking for full-time AI/ML/LLM engineering, applied AI, agent, research-engineering, forward-deployed, and ML systems opportunities starting around my December 2026 graduation.