// Staff AI Engineer
Shipping production AI systems end-to-end.
11+ years shipping complete production systems including LLM, RAG, agentic AI, and distributed platforms delivering 25–70% efficiency gains at 200+ Gbps scale across aviation, autonomous vehicles, robotics, and enterprise. IEEE-published researcher. Hands-on with Python, LangGraph, LangChain, and MCP.
Staff-level AI/software engineer with 11+ years shipping complete production systems end-to-end, from founding and principal engineering roles into Staff-level technical ownership.
My work spans LLM, RAG, agentic AI, distributed systems, and real-time platforms delivering 25–70% efficiency gains at 200+ Gbps scale across aviation (Etihad Airways), autonomous vehicles (Rivian/VW, Applied Intuition), warehouse robotics (Berkshire Grey), and precision agriculture (Verdant Robotics).
I combine deep software engineering fundamentals with applied AI, with hands-on expertise in Python, LLMs, RAG, MCP, LangGraph, and LangChain. I have a track record of making architectural tradeoffs, productionizing non-deterministic AI systems, and owning outcomes from architecture through deployment. My research on pedestrian intent prediction and human tracking has been cited 50+ times.
Autonomous multi-step research agent built with LangGraph. Plans a research strategy, fans out parallel web searches, extracts and synthesizes findings, and produces a cited markdown report without human intervention.
A collection of production-ready MCP (Model Context Protocol) tool servers for AI agents — web search, HTTP fetch, file ops, SQLite queries, sandboxed code execution, and persistent agent memory.
Flexible RAG pipeline that ingests PDFs, web pages, markdown, and YouTube transcripts, indexes them with hybrid retrieval, and answers questions with cited responses. CLI and FastAPI interface included.
LLM agents argue opposing sides of any topic across multiple rounds, then a judge agent scores both sides on factual accuracy, logic, and persuasiveness and declares a winner with reasoning.
Lightweight evaluation toolkit for LLM applications. Measures faithfulness, answer relevance, context precision, and semantic drift. Runs as a pytest suite in CI and generates HTML reports with pass/fail verdicts.
A minimal, complete MLOps setup wiring experiment tracking, model registry, containerized serving, automated validation, and CI/CD together. A ready-to-clone starting point for any production ML project.
Hand gesture mimicking system and course notebooks built for the AI4ALL program at University of Maryland. Explores human-robot interaction through real-time gesture recognition.
Cooperative multi-robot system built for the NASA Swarmathon challenge. Robots collaboratively search and collect resources using swarm intelligence, SLAM, and autonomous navigation.
Detects and classifies traffic signs from images using MSER feature extraction and an SVM classifier. Outputs bounding boxes around detected signs with class labels.
Estimates the trajectory of a camera mounted inside a moving vehicle using sequential frame analysis. Compares a custom implementation against OpenCV's built-in visual odometry pipeline.
Implements the Lucas-Kanade optical flow algorithm to track objects (car, human, vase) across video frames using template matching and iterative gradient descent.
Continuously tracks three differently colored underwater buoys in real-time video using color segmentation and contour analysis in challenging underwater lighting conditions.
Detects and tracks road lanes from a camera mounted on an autonomous vehicle. Uses Hough transforms and sliding window analysis, with turn prediction from lane curvature.
Detects and tracks AR tags in video frames and overlays 3D objects onto them in real time. Uses homography estimation and perspective projection for accurate augmented reality rendering.
Novel reinforcement learning approach to robot relocation for improved human instance segmentation and tracking, with applications in autonomous navigation, social robots, and human-robot interaction.
Emotion-aware neural network for predicting pedestrian crossing intent, fusing appearance and behavioral cues for safer and more natural autonomous social robot navigation.
Open to senior AI engineering roles, advisory engagements, and research collaborations. Let's build something intelligent.