Things I've built — AI products, web apps, research tools, and side experiments. All shipped, most still running.
Clinical RAG system for smoking cessation research at UTSA Mobile Lab. Grounded retrieval over peer-reviewed literature with a 0.72 cosine refusal threshold. Published in Health Education Research (Q1).
Global earthquake detection & tsunami warning network. Real-time seismic monitoring for the UTSA research team — per-client queue + reconnect-aware delivery, 2s average detection latency, zero message loss.
Persistent memory for Claude Desktop — semantic search over documents and conversations with grounded answers and citations. Sub-10ms ingestion latency. MCP server with 10 tools, 5 resources. MIT licensed.
Messaging protocol engine for health intervention studies. Design branching message sequences, schedule timezone-aware delivery, monitor queue health. Built for research teams running intervention studies at scale — powers the QuitTxt 2025 protocol.
Pluggable web extraction infrastructure with automatic engine fallback. One REST API across CrawlAI, Scrapy, and Firecrawl — 50+ production sites, 95% handled without manual intervention.
Your WHOOP data, decoded. Precision health dashboard with 7-day trends, training readiness scores, and AI-derived insights. Per-user OAuth vault, real-time freshness, color-coded data signals.
Multi-agent coordination system built for a startup — autonomous task delegation, tool use, and shared memory across agents. Handles 500+ concurrent sessions in production with zero message loss.
Multimodal foundation model training framework — unified training pipeline across text, image, and audio modalities. Built for experimenting with cross-modal pretraining strategies.
Full-stack application management tool with status pipelines, analytics, and reminders. Supabase backend + Next.js frontend.
Real-time quantitative analytics platform for financial data processing. ML pipeline, REST API, and visualization layer — designed to handle streaming market data at scale.
Predictive hotspot detection system for Austin traffic incidents. Enables proactive emergency response positioning by forecasting high-risk areas before incidents occur.
Full-stack app for exploring Rig Veda texts — NLP-powered search, verse navigation, and commentary features. End-to-end: FastAPI backend, React frontend, custom text parsing pipeline.