RAG-powered AI voice assistant for emergency responders. Sounds human, and takes input by reading emotional intent and situational context. RAG over Safety Datasheets. 97% retrieval accuracy on queries by chemical trade name and scenario. Took it from POC to MVP.

Document data extraction service. Cut $2,000 to $3,000 per 1,000 pages down to $30 to $50. Brought processing from two or three weeks down to ten minutes. 98%+ SDS accuracy, at or above manual labor. Now being productionized with VLMs and pgvector for the flagship.

This site. Next.js 15 with React 19, statically exported and served on Firebase Hosting. GitHub Actions pipeline lints, builds, and deploys on every push to main. Animated orbit-of-interests constellation on the home page. Single accent color, oversized grotesque headlines, custom SVG illustrations for every project.

Senior design (CSCE 483) for Texas A&M Formula SAE Electric. On-car telemetry display plus a pit-side web app for live monitoring and configuration. Built the React frontend, contributed to the Express/Firebase backend, and owned the cloud system design. Firebase auth, WebSocket streaming, Fly.io.

Co-authored research paper for MATH 414 (Fourier Wavelets) at Texas A&M. Wavelet-based singularity detection in ECG signals across five wavelets and two pipelines (CWT and DWT). Best F1 0.907 on QRS detection against the MIT-BIH Arrhythmia Database.

Personal project for CSCE 642 (Deep Reinforcement Learning) at Texas A&M. DRL system that picks the right LLM per prompt, like the 'auto' model selector in coding agents, balancing cost and capability per task. Five agents (DQN, PPO, LinUCB, PickLLM, Greedy) trained on RouterBench, with feature-based and embedding-based variants and a Streamlit evaluation dashboard.

CSCE 431 (Software Engineering) project from my Singapore study abroad, built for the Texas A&M Engineering Department. Ruby on Rails platform with 40 templates across 10 topics, generating unlimited practice variants. Team of 6, faculty client.

Drug-candidate research platform for biotech sponsor NeuroX1. Hybrid retrieval and RAG through LangChain (LLM and BioMixtral 8x22B). 5x faster data access and 62% lower cost per employee hour. The sponsor took it on after delivery.

Five full-stack AI projects in three months, one not public. LangChain, Pinecone, AWS, GCP, computer vision. Mentored by engineers from Google, Meta, Apple, and DoorDash, plus founders and researchers from Stanford and Berkeley.

Sole developer on a full-stack platform to remotely manage a 3D-printer farm in Belgium. Took the team from a handful of printers running in parallel to seven times that. React, Node, Python, Firebase.