Charles Feinn

Spearheaded multi-language code review and delivery, debugging and verifying production code across Python, TypeScript, C#, and Rust for large codebases, ensuring end-to-end software shipment.

Developed and open-sourced an LLM agent harness with a comprehensive evaluation suite, featuring code-based assertions across 10 capability categories, a three-tier citation-verification pipeline, and deployed as a live demo.

Reproduced and enhanced a published LLM benchmark, achieving a 71.1 MMLU macro score (+2.8 points over manual prompting) through automated DSPy-based prompt optimization, and published the methodology.

Architected and implemented end-to-end AI and backend systems, including a multi-tenant FastAPI/PostgreSQL service with row-level-security and optimistic concurrency, alongside a multi-provider LLM server featuring circuit breakers, pipeline processing, and CI/CD.

Provided technical leadership for the mobile team, selected by the outgoing lead due to expertise in debugging and reviewing a large, mature codebase across production applications.

Architected a shared cross-platform framework, standardizing authentication, UI, and media playback, reducing new-app development time by ~50% and contributing to a 30% reduction in customer churn.

Established CI/CD pipelines (Fastlane + GitHub Actions) and performance monitoring, decreasing downtime from days to hours across the team's projects.

Designed and integrated an ML recommendation system (content-based + collaborative filtering), scaling the platform to 24,000 monthly active users.

Directed R&D efforts on content-moderation classifiers, ensuring platform integrity and user safety standards.

Successfully founded and scaled a music-technology startup, growing the user base to over 1 million users.

Led a team of 10 professionals across product development, fundraising, and engineering, demonstrating extensive self-directed ownership of complex technical problems.

Secured funding and managed engineering initiatives to drive platform growth and user engagement.