Ryan Tolone
Hi, I’m Ryan — I build deployable intelligence that ships.
UCLA statistics & data science senior obsessed with translating research-grade AI into reliable, product-ready systems for sports vision, nutrition coaching, and algorithmic trading.
applied ml engineering full-stack data products responsible ai
What I’m optimising right now
I’m happiest when I can combine rapid experimentation with production discipline—owning the data plumbing, models, evaluation, and product polish in one loop.
Data Systems that stay trustworthy
Designing drift-aware monitoring for RAG stacks, instrumenting feature stores, and keeping evals tightly coupled to user intent.
Products with measurable lift
Translating ML into growth metrics—think CV pipelines for sports analytics, nutrition adherence scores, and trading P&L guardrails.
Full-stack execution
Wireframing UX, building ETL in the cloud, shipping inference APIs, and closing the loop with storytelling dashboards.
Highlighted builds
BiteBuddy AI
Vision-powered meal logging plus retrieval-augmented nutrition coaching. Led everything from on-device inference to pilot cohort activation.
Pickleball vision stack
Automated rally segmentation, spin tracking, and win probability overlays for competitive pickleball using multi-view pose estimation.
Algorithmic trading engine
Deployed an ORB-inspired equities strategy with live risk guardrails, Azure Functions scheduling, and discord-based anomaly surfacing.
Recent trajectory
2025 • BiteBuddy AI Leading a second pilot with UCLA club athletes, expanding the recommendation engine with retrieval-augmented meal plans.
2024 • APE Picks Vision Lab Built the computer vision analysis pipeline powering collegiate pickleball scouting reports and live broadcast overlays.
2024 • Algorithmic Trading Collective Architected data ingestion, feature engineering, and execution for a community-driven intraday trading strategy.
Latest writing
I publish deep dives on infrastructure that keeps AI honest—from LLM drift guardrails to federated feature stores.
Monitoring LLM Drift with Synthetic Control Charts
Why retrieval-augmented apps need streaming semantic telemetry and how to wire probe cohorts that actually detect drift.
Beyond Specialists — Multi-Game Training
Exploring generalized policies for incomplete information games and the implications for competitive poker AI.
Federated Feature Stores in Digital Health
How edge-personalized models can deliver cardiology insights in real time without compromising HIPAA compliance.
Let’s collaborate
I’m open to internships, research collaborations, and product experiments in responsible AI, data infra, or sports analytics. Reach out and let’s build something measurable.