Work
Writing, research, and projects.
Tests Turn Prompting into Search
What I learned building the evaluation pipeline behind BODE-GEN, where executable tests turn code prompting into an optimization problem.
ScenarioLens
An autonomy-evaluation tool for ranking and inspecting long-tail driving scenarios with interpretable risk and baseline-failure metrics.
MetricDrive
A research preview for studying metric-aligned trajectory choices, preference pairs, hard negatives, and VLM-style planning rows.
BODE-GEN
Bayesian prompt optimization for test-driven code generation, evaluated across HumanEval+ tasks and multiple code models.Self-Driving Car
An end-to-end behavioral-cloning system that predicts steering from camera images and drives in the Udacity simulator.IntelliCrawl
A source-backed research agent that discovers, profiles, and compares developer tools while preserving field-level evidence and explicit unknowns.SentiSync
A YouTube sentiment-analysis extension backed by an MLflow, DVC, Docker, and AWS machine-learning workflow.Knowledge Graph RAG Assistant
For this team-built knowledge-graph RAG assistant, I generated and indexed 5GB+ of Wikipedia embeddings and refined FAISS retrieval toward a 10,000+ article knowledge base.
Social Triangles
An IROS study of how multi-robot formation geometry shapes human navigation and approach behavior.