MonsterGPT
AI betting copilot that merges projection models, live sportsbook odds, and RAG-based web research into tool-driven answers and bet suggestions.
Carson Szeder
Software engineer & founder.
I build real products for real users: prediction models, market infrastructure, and AI agents that run in production 24/7. Featured in WIRED for applying AI to sports betting.
Core expertise in agentic systems — the RAG pipelines, knowledge graphs, and open-source model stacks that make AI agents actually work in production.
02 — Work
Founder & Lead Software Engineer
- Built the business from zero to one — architecture, product, data, and infrastructure — operating as founding engineer on every layer.
- Designed and shipped a production Rust API (70+ endpoints, WebSocket streaming) and the Next.js / React customer app (270+ components, typed clients, real-time charts, Ethereum & Solana wallet support).
- Built a real-time data platform covering 13 sports, 50+ sportsbooks, and prediction markets (Polymarket, Kalshi, Novig, ProphetX) — 100+ continuously running services collecting odds, results, and order-book data around the clock.
- Developed pricing and prediction models for moneylines, spreads, totals, player props, DFS, and arbitrage — removing bookmaker margin and modeling exchange liquidity.
- Built MonsterGPT, an AI betting copilot combining internal projections and live market data with retrieval-augmented web research — answering questions and surfacing bets through tool calls, not static prompts.
- Shipped low-latency HFT execution in Rust: Kalshi parlay RFQ market-making with live order-book rechecks, exposure limits, and full audit trails — plus a Polymarket pipeline that provisions user wallets, syncs allowances, and executes and reconciles orders with real-time risk checks.
03 — Selected projects
Things I've built
Classroom AI
Consumer AI schoolwork assistant launched right after ChatGPT released. Reached 25,000 users.
Kalshi RFQ Market Maker
High-frequency, low-latency Rust service that prices multi-leg parlay RFQs from live order books — rechecks book state pre-trade, models fees and slippage, caps exposure, prevents duplicate orders race-free, and submits signed quotes with full audit logging.
AI Content Agents
Agentic content pipeline — agents research competitors and audience trends, draft hooks and scripts, and generate image and video creative through Higgsfield MCP. Automated Meta Ads upload, review, and publishing; Airbyte pipelines land performance data into a self-hosted ClickHouse warehouse.
Yard AI
Baseball tracking and analytics platform that organized player and team data so coaches and players could actually review performance.
Film Discovery Tools
LLM tools that turn natural-language requests into structured filters, retrieve relevant titles, and explain recommendations against your criteria.
04 — Skills
What I work with
Languages
- Python
- Rust
- TypeScript · JavaScript
- SQL · HTML · CSS
AI — Agentic
- LLM tool use & RAG
- Knowledge graphs
- Agentic orchestration
- MCP tool integration
- Context engineering
AI — Open Models
- Open-weight & uncensored builds
- vLLM · SGLang · llama.cpp
- EXL3 quants · spec decoding
- Multi-GPU serving · TP4
- LightGBM · XGBoost · sklearn
Data & Backend
- PostgreSQL · Supabase
- Self-hosted ClickHouse
- ETL pipelines · scraping
- Actix-web · Tokio
- REST APIs · WebSockets
Frontend & Infra
- Next.js · React
- React Native · Tailwind
- Zustand · TanStack Query
- Railway · DigitalOcean
- PM2 · systemd · Git
05 — Off the field
Former college baseball player
Left-handed pitcher at Folsom Lake College — competed in two California State Championships and took the win in a championship final. Also once ranked among the top 1,000 Fortnite players in the US.
I'm intensely competitive, and I like building products that hold up under real use.
Looking for software?
If you're looking to get software built, let's talk.
I'm open to engineering work and new projects. Email me directly and I'll get back to you.
carson@whitelotustechnology.com →