Applied AI / eval / agent systems · sports decision products · Alloway LLC.
Website: ianalloway.xyz · Writing: Alloway AI
I build and evaluate systems where AI products, agent tooling, and decision software have to hold up under scrutiny.
- Contract @ Handshake AI — evaluating ChatGPT app development and agent-built software across engineering, logic, design, and product quality
- Founder, Alloway LLC — applied ML products, analytics tooling, and autonomous agent systems (including AI Advantage Sports)
- B.S. Information Science, magna cum laude · M.S. Artificial Intelligence candidate — University of South Florida
I care about things that still hold up when someone looks closely & over time.
Case study — AI Advantage (+ SOLVENT)
Problem → system design → what “good” means → honest results. Live product + repos. No fabricated metrics.
SOLVENT — Self-Funding AI Agent
NVIDIA × Stripe × Nous Research Hackathon
An AI agent that runs as a business loop: sells research briefs, collects payment via Stripe, provisions compute from its own revenue, and refuses jobs that don't clear a margin. Demo runs book a full P&L (earn → fulfil → spend); dollar figures in the demo are illustrative of the loop, not production revenue.
Client pays Stripe → Agent earns → Agent fulfils → Agent pays vendors → P&L booked
Sponsor stack: NVIDIA Nemotron · Stripe Issuing · NemoClaw-style guardrails.
Zero-dependency Python toolkit for auditing LLM-as-judge pipelines — position / verbosity / injection bias checks, judge panels with agreement scoring, and calibration (Brier, ECE) against human labels. pip install juryrig
Full-stack sports analytics product with live ML predictions, Kelly-based sizing, and a real deployed surface at aiadvantagesports.com.
Stack layering: nba-ratings (PyPI: nba-edge) → kelly-js (npm install @ianalloway/kelly-js) → sports-betting-ml (training demo) → ai-advantage (product).
| Repo | One-liner |
|---|---|
| nba-ratings · PyPI | Elo, win probability, calibration — nba-edge on PyPI |
| juryrig · PyPI | LLM-as-judge audit toolkit — pip install juryrig |
| kelly-js · npm | Kelly sizing, CLV, bankroll — npm install @ianalloway/kelly-js |
| openclaw-skills | 15 OpenClaw/ClawHub agent skills (sports-odds, Kelly, portfolio, …) |
| sports-betting-ml | Streamlit training / value-bet demo (synthetic demo metrics) |
Eval notes that still matter: browser automation remains a hard agent benchmark — Substack · checklist.
More: ianalloway.xyz
Languages: Python · TypeScript · SQL · JavaScript
AI / ML: OpenAI · NVIDIA Nemotron · LangChain · scikit-learn · Pandas
Product: FastAPI · Next.js / React · Stripe · PostgreSQL · Docker · GitHub Actions
- Autonomous agent systems with real economic loops (earn, spend, book P&L)
- Evaluation tooling that makes model and app behavior inspectable and trustworthy
- Applied ML products where the evaluation layer is part of the system, not an afterthought
- Sports decision systems — ratings → edge → Kelly → live product (AI Advantage)
other agents can spend money · SOLVENT gates spend on margin
