Projects

Selected work in AI, machine learning, data systems, and product engineering.

Industry-University Collaboration with COSCO SHIPPING Technology

View details

Led development of a commodity-trading intelligence platform that integrates vessel, port, freight-rate, and arrival data with FastAPI and PostgreSQL. Built LangGraph agents for NL2SQL analytics, sentiment analysis, anomaly alerts, and maritime intelligence across AIS, radar, permit, enforcement, port, and S-100 geospatial data.

#FastAPI#PostgreSQL#LangGraph#NL2SQL#Maritime Data

Horizon: Personalized AI News Radar

View details

Contributed an asynchronous Python workflow for multi-source ingestion, deduplication, LLM importance scoring, summarization, and webhook delivery. Designed the Balanced Digest selection strategy, combining LLM scores with category quotas to improve coverage and reduce model calls.

#Python#Async Workflows#LLM#Webhooks
View repository

LLM Polysemanticity Study

View details

Designed a Pythia-6.9B evaluation framework using causal tracing and collision analysis to locate polysemantic neurons and compare the collateral damage of neuron-level FiNE editing with layer-level ROME editing.

#PyTorch#Model Editing#Causal Tracing#Transformers
View repository

Ranno.ai: Safe Route Agent

View details

Built an on-device AI agent with Chrome Prompt API and Gemini Nano to parse natural-language requests into structured intents, then combined Google Maps with San Francisco crime and municipal amenity data to generate personalized, safety-aware routes.

#Gemini Nano#Chrome Prompt API#Google Maps API#TypeScript
View repository