Brandon Han

Building an AI agent safety engine that uses Thompson Sampling to learn safe behavior from human corrections. Designed the core decision gate SDK (check, override, status) and built the full-stack platform — Python/FastAPI backend, React dashboard, PostgreSQL data layer, and a published Python SDK — as a solo technical founder.

Architected a behavioral learning engine with per-tool-category Beta distributions that outperforms static rule-based guardrails after ~50 human override signals. Extracted and generalized the algorithm from a production personal finance app (50+ ML features, HistGradientBoosting risk scoring, multi-tenant SaaS).

Conducted competitive analysis of the AI agent guardrails market (Patronus AI, Virtue AI, Straiker, Guardrails AI) and positioned the product as a developer-first agent safety platform targeting startups and SMBs deploying autonomous AI agents.

Defined the product vision and roadmap for an AI-powered iOS budgeting app, taking it from idea through MVP launch with real users connecting actual bank accounts via Plaid.

Built a transaction pipeline that ingests live bank data, categorizes spending automatically, and surfaces personalized budget recommendations — replacing what most people try to do in a spreadsheet. Ran competitive analysis against YNAB, Monarch, and Mint to identify underserved segments, then used those findings to prioritize features and refine onboarding.

Live on the App Store with 300+ waitlist signups pre-launch. Android in development.

Owned the end-to-end rollout of 100+ school websites and apps for Los Angeles Unified School District — a $10M+ engagement — managing timelines, resolving blockers, and coordinating across teams and district stakeholders.

Built dashboards from raw SQL data that surfaced rollout bottlenecks and helped prioritize which schools to launch next. Created Python automation scripts that eliminated repetitive workflows, saving 40+ hours of manual work and over $10K across multiple projects.