Statistics & Data Science @ UT Austin. I do research and build across the ML stack, from computer vision pipelines to retrieval systems to full-stack AI products. I care about systems that are measured, not just demoed: benchmarked retrieval, precision-tuned detection, and LLM features with guardrails against hallucination.

Interests: RAG · computer vision · full-stack AI · applied machine learning

Section-aware retrieval, citation-graph expansion, multi-hop reasoning, and NLI-based claim verification over NLP/ML papers. Every component is benchmarked individually against QASPER and SciFact. Python · RAG · LangGraph · SPECTER2 · Qdrant

Multi-source ingestion + LLM summarization, with a two-tower neural recommender trained via contrastive learning and Thompson-sampling exploration to avoid filter bubbles. PyTorch · FastAPI · Recommender systems · LLM

Two-stage CV pipeline (YOLO + ByteTrack detection feeding a ResNet18 classifier) that finds Waymo vehicles in street video and exports clips. Tuned for precision over recall. Research at INSPIRE Lab, UT Austin. Computer Vision · PyTorch · YOLO · OpenCV

A chat-first workspace with a persona-aware assistant: native Anthropic tool-use loop, a versioned wiki, grounded retrieval, scheduled agents, and an anti-hallucination guardrail on LLM extraction. Next.js · TypeScript · Prisma · PostgreSQL · Tool-use

A Claude-powered assistant that runs your day over text: morning briefings, smart reminders, deep-work focus sessions, location-triggered fitness logging, and conversational nightly reviews, all on a cron + webhook backbone. Next.js · TypeScript · Supabase · Twilio · Claude Agent SDK

Languages: Python · TypeScript · SQL ML / AI: PyTorch · Hugging Face · LangGraph · RAG · embeddings · recommender systems Full-stack: Next.js · FastAPI · PostgreSQL · Prisma Infra: Docker · Qdrant · Alembic

📫 Reach me through GitHub or the links on my profile.