I'm a software/AI/ML engineer currently studying computer science at Purdue University. I love exploring new technologies, reading research papers, and building cool projects.
- 💼 Previously: SDE Intern at AWS, AI SWE Intern at Nokia, and AI Research Engineer at ML@Purdue x Captain (YC W26)
- 🏆 Winner of the 2026 Claude Hackathon @ Purdue with Room Reveal
- 🔭 Currently building Triaxis, a community-powered map of competitive games
Small things I built and use pretty much every day.
| Tool | What it does |
|---|---|
| mdview | Terminal-launched Markdown viewer with vim keybindings, live reload, KaTeX math and GitHub-flavored callouts. Opens a chromeless browser window (no Electron), and works natively from WSL. |
| ed-discussion-agent | Agent tooling for Ed Discussion. A read-only MCP server lets Claude Code, Codex, Cursor, and other MCP clients search and read your courses, and an experimental listener pushes notifications to your phone (like "tell me if any CS 180 deadline changes"), judged by a local System One decision model. On PyPI: run it with uvx ed-discussion-mcp. |
| claude-study-skills | Claude Code plugin with three skills: ingest a course's slides and transcripts, generate printable exam reference sheets with parallel subagents, and build study guides that teach a topic. |
| dining-bot | CLI, TUI and agent skill for Purdue's dining menus. npx skills add GalacticQuasar/dining-bot, then ask "What are some good vegetarian options for dinner tonight?" |
| prose-diff | Merge two versions of your writing one change at a time, with word-level inline diffs, accept/reject per change and vim-style shortcuts. Paste two versions into the web app, which runs fully in the browser, or run npx prose-diff draft.md, let an agent like Claude Code edit the file, then review its edits and write the merged result back. Live · npm |
| prompt-builder | PWA for local, browser-based prompt engineering. Live |
| task-graph | To-do list modeled as a forest of task sequences that surfaces only the actionable frontier: the tasks you can do right now. Graph view for the whole forest, plus quick capture from your phone. |
| Project | What it is | Built with |
|---|---|---|
| Room Reveal | Crowdsourced interactive 3D models of Purdue dorm rooms, so students (especially those with accessibility needs) can see a space before moving in. Live · Devpost | Gaussian Splatting, Nerfstudio, Three.js, FastAPI, Modal |
| DeepQuery.org | Open-source arena-based RAG benchmark ranking pipelines by accuracy, speed and cost with LLM-as-a-Judge ELO. Designed and developed the core LLM-as-a-Judge ELO evaluation pipeline, built RAG pipelines with cross-encoder reranking, and co-authored the paper. | LangChain, RAG, LLM-as-a-Judge |
| machine(learn); | Agent swarm that plans, writes, trains and tunes PyTorch models on parallel serverless GPUs (Catapult 2026). | vLLM, Modal, Supabase, Next.js |
| Clariti | Voice-first memory companion for people living with dementia (HackIllinois 2026). | Multimodal RAG, InsightFace, Expo, FastAPI, ElevenLabs |
| Nova LLM | Full-stack LLM agent with custom tool calling and configurable MCP servers. Live | React, Express, MCP, GenAI SDK |
More experiments
- tombstone-agent: a controlled study of how agents can forget their own tool results, compared against truncation and auto-compaction
- ai-research-assistant: search arXiv, keep a local paper library, get structured summaries and chat with each paper
- system-one-testing: experiments with a tiny local 0.8B model via Ollama: support-ticket triage, a dino runner, a rule-changing arcade game and a car game
- face-clustering: cluster faces in a photo collection by person using DeepFace/FaceNet embeddings and DBSCAN
- ed-graph: build and visualize a graph of how Ed Discussion threads reference each other
- FacialRecognition: face detection and recognition pipelines for images and live video
- ThyroidDiseaseClassifier: my StARLinG Lab @ UT Dallas research
- OCRType: screenshot to simulated typing with OCR



