Scaled self-architected platform to over 14,000 monthly users and 2.5 million ad impressions, maintaining 99.9% uptime on a single VPS. Engineered the end-to-end architecture (frontend, backend, database), slashing operational costs by 80% by deploying self-managed Docker containers. Orchestrated the PostgreSQL and Redis infrastructure within a monorepo, optimizing the CI/CD build pipeline to cut deployment times by 40%.

Enhanced LLM code generation accuracy by 30% through rigorous debugging and validation of TypeScript and Python outputs. Evaluated model outputs across critical axes including truthfulness, safety, and instruction following, contributing directly to RLHF (Reinforcement Learning from Human Feedback) pipelines. Designed complex system prompts and adversarial test cases to rigorously test model capabilities and reduce AI hallucinations.

Maintained and expanded a client e-commerce application using Next.js, implementing responsive UI components based on precise design requirements. Developed 5+ REST API endpoints to handle product data, successfully bridging the frontend client with the backend logic.

Developed core frontend modules for a Media Asset Management (MAM) system using React, enabling users to efficiently upload, organize, and retrieve digital assets. Integrated UI components with a Python backend, utilizing AWS S3 for file storage and DynamoDB for metadata management. Delivered 15+ major feature updates, including advanced search filters and file previewers, ensuring a seamless user experience for managing complex data.