Co-founded Fargo Automotive of Gainesville in 2018 and ran it for 7+ years — built and led the tech end-to-end, scaled to 5M+ daily requests, 99.9% uptime. Now transitioning full-time to AI: shipped 5 production AI products in 2025–26, with a paper under review at AIES 2026 on safety-critical conversational AI. Open to YC funding and AI startup roles where I can build product zero-to-one.
Production U.S. real estate analysis platform. Trained a gradient-boosted AVM that hits 18.6% MAPE in NYC and 15.2% MAPE in Connecticut on holdout data, paired with a 6-agent Claude-powered analysis pipeline (school / commute / climate / cost trajectory / neighborhood / valuation). Full GA4 + Search Console funnel instrumentation; deployed on GCP Cloud Run + Cloud SQL behind Cloudflare.
Open-source mental-health companion built on Next.js + Claude with an explicit crisis-detection layer (988 / Crisis Text Line / DV / Childhelp bridging) and a clinician-reviewed safety protocol. First-author on two papers submitted to AIES 2026 on safety-critical conversational AI; informal clinician review completed; prompt + skill distributed under a custom safety license.
Full-stack consumer AI app with an 8-agent reasoning pipeline and a novel /ritual gesture flow that uses the device camera for an interactive divination experience. Migrated the production stack off Render onto GCP Cloud Run + Cloud SQL fronted by a Cloudflare Worker, with a pre-compute strategy that cut repeat-reading inference cost by ~83%.
End-to-end automation that pulls YouTube videos, transcribes & dubs them into Chinese with cloned voices, and uploads to multiple Chinese platforms. Operates an 18-channel YouTube matrix on automated systemd timers (08 / 14 / 20 uploads + 09 Telegram daily report), with multi-account session management and quota-aware scheduling.
Self-hosted AI animation pipeline turning Chinese wuxia novels into animated short dramas. Wires together SkyReels-V2 for video, CosyVoice for voice cloning, and Kling for shot-level retries — all running locally on a single RTX 5090 with a custom GPU broker (MCP) for VRAM reservation across concurrent jobs.
Open to opportunities in cloud architecture, distributed systems, and AI/ML engineering.