← Back to experience

Agents

LLM-powered systems, RAG pipelines, and intelligent tooling

Lead Developer @ Open Project @ Berkeley

Jan 2025 — May 2025
  • Built an LLM-powered RAG system for chess coaching — improved agent performance by 20%
  • Extracted tactical and strategic features from board positions and retrieved similar historical positions and concept templates
  • Developed an offline evaluation pipeline over 2M+ positions to test multiple LLMs with and without generated hints
PythonReact NativeFlaskSQLite

Machine Learning Engineer @ NeuroTech @ Berkeley

Jan 2026 — Present
  • Built an end-to-end BCI pipeline translating 16-channel EEG signals into real-time gait angle predictions for exoskeleton control
  • Designed the preprocessing pipeline with artifact removal, bandpass filtering, and sliding window alignment for streaming inference
  • Achieved 5° RMSE enabling responsive, closed-loop control between brain signals and exoskeleton actuation
PythonPyTorchEEG/BCIPipeline

Projects

Structured Document Editing Agent

LLM-powered agent for structured document editing via tool calls. Uses an intermediate representation for safe, consistent multi-step edits — reducing manual editing time by 60%.

TypeScriptAI SDKLLM Agents

AI File Organizer

AI-based tool using recursive traversal, semantic indexing, and model-guided planning to organize directory trees. Two-stage Explorer/Organizer pipeline achieving 18x faster cleanup.

RustSemantic AIPipeline