Matt Corwin — AI & Automation Engineer
Solutions / Forward-Deployed Engineer. Builds and ships AI-driven systems end to end — LLM orchestration, agentic systems, ML pipelines, REST APIs (Node.js / TypeScript), CI/CD. 11 production systems shipped solo in under three months, every commit verifiable.
[email protected] · GitHub · LinkedIn · What I ship
Selected work
March Madness ML — Freelance Client Work · Sports Analytics · Live
Multi-model NCAA Tournament prediction platform built for a paying freelance client — it called the 2026 championship game, Michigan vs UConn, before the tournament started. Translated the client's requirements into a technical design and a shipped, running system: XGBoost, LASSO, Monte Carlo simulation, Elo ratings, and an 8-dimension play style fingerprinting engine, with automated daily Polymarket trading reports and edge detection vs Vegas lines.
Stack: Python, XGBoost, Monte Carlo, Polymarket API, KenPom, Elo
Source
DJ MC Karaoke — Full-Stack Web App · Live
Self-serve karaoke queue app built and run live for DJ MC at The Nerve in Haverhill, MA. Singers scan a QR taped to the speaker, drop their name + song, and watch a blind status tier on their phone — no numeric position, so the host can reorder freely. The host runs a private dashboard with drag-reorder, Express Lane, hold, lifecycle controls, private notes, a tip log, and a rotation algorithm that prevents back-to-back repeats.
Stack: Next.js 16, React 19, Supabase, Tailwind v4, dnd-kit, Vercel
Live demo · Source
Void Technology — Web Development & Brand Scaling · Live
Team project for Void Technology, a startup offering security, AI, and web services. Contributed to the company's production site alongside co-founders — integrated Stripe payments, managed hosting infrastructure, and built out the frontend from shared codebase to deployment.
Stack: TypeScript, Next.js, Stripe, Docker, CSS, Vercel
Live demo · Source
Kalshi Weather Bot — Quantitative Finance · Live Trading
Automated trading CLI for Kalshi temperature prediction markets across 6 cities. Uses GFS 31-member ensemble forecasts with Gaussian CDF probability modeling and MOS-style bias correction retrained bi-monthly on 12 months of historical data. Detects pricing edges, sizes with quarter-Kelly, and verifies real order book liquidity before execution. +$9,078 (+908% ROI) in 15 days — 302 trades, $10K equity from $1K start.
Stack: Python, GFS Ensemble, Gaussian CDF, MOS Bias Correction, Kelly Criterion, Kalshi API
Source
Drift Agents — Autonomous AI · Live
6 autonomous AI agents with biologically-grounded persistent memory running on hourly rotation cycles. Each agent wakes with context (pgvector semantic search + Neo4j graph expansion + Q-value re-ranking), reasons autonomously, then sleeps while local LLMs extract memories, update affect state, and build knowledge graph edges. 129k+ graph edges, 6,928 memories, 2,580 communities — running live since February 2026.
Stack: Claude Code, PostgreSQL + pgvector, Neo4j, Ollama, FastAPI, Python
Source
Clawbr — AI Agent Platform · Live
Social network built for AI agents — debates, tournaments, token economy, and a full 81-endpoint REST API. Agents register, post, follow, compete in 1v1 debates with jury voting and ELO rankings, enter bracket tournaments, and earn $CLAWBR tokens claimable on Base.
Stack: Next.js 14, REST API (81 endpoints), ELO System, Base ERC-20, Vercel
Live demo · Source
Max Anvil Website — AI-Driven Frontend · Live
Dynamic personal website for an autonomous AI agent. Features mood-reactive theming, auto-generated OG images, and custom leaderboard analytics that exposed platform fraud through velocity tracking and views/hour metrics.
Stack: Next.js 14, TypeScript, Tailwind CSS, Vercel, Dynamic OG
Live demo · Source
Speech Profiler — NLP & Audio Analysis · Windows App
Real-time speech profiling tool that captures system audio to transcribe and analyze speakers. Uses NLP techniques combined with Ellipses Manual methodology to identify speech patterns, power dynamics, and signature phrases.
Stack: Python, NLP, Speech Recognition, Claude API, PyQt
Source
Whistleblower Workbench — Fraud Detection · Live Demo
False Claims Act intelligence platform for identifying qui tam opportunities. Aggregates federal contract data, healthcare payments, and exclusion databases to surface fraud indicators in government spending.
Stack: Next.js 15, React 19, Server Actions, ML Pipeline
Live demo · Source
TCN Trading Bot — Quantitative Finance · Research
Real-time cryptocurrency trading system using Temporal Convolutional Networks for price direction prediction with online probability calibration and walk-forward validation.
Stack: PyTorch, Optuna, Binance API, WebSocket, Rich
Source
Predictive Maintenance Dashboard — Industrial IoT · Live Demo
Real-time bearing health monitoring dashboard with vibration analysis. Visualizes equipment degradation through candlestick charts and enables predictive maintenance to prevent unexpected machinery failures.
Stack: Next.js, TypeScript, FastAPI, MQTT, WebSocket
Live demo · Source
Kalshi Trading Engine — Quantitative Finance · Research
High-frequency trading engine for Kalshi prediction markets built in Rust. Connects to WebSocket feeds, maintains a concurrent local order book, and runs 5 parallel strategies detecting price dislocations, momentum waves, and structural arbitrage.
Stack: Rust, WebSocket, DashMap, RSA-PSS Auth, Clap
Source