Aug 2025–Present
KVA Platform
Full-Stack & AI Engineer
Built the KVA platform end-to-end — website, design system, internal infrastructure, custom OAuth2, RAG pipeline with vector embeddings and LLM integration.
KVA is an agentic development platform where AI agents autonomously execute software tasks on real GitHub repositories. I joined as the founding engineer and built everything from the ground up — from the public-facing website and design system to the core infrastructure that powers AI agent execution.
The RAG pipeline ingests project documentation, codebase context, and engineering notes, generates vector embeddings via OpenAI, and retrieves relevant context at query time to ground LLM responses. Custom OAuth2 handles authentication across services. BullMQ + Redis manages the distributed job queue for agent task orchestration.
The architecture is designed for reliability at scale: each agent run is a stateful job with retries, timeouts, and structured logging. PostgreSQL stores task state, execution history, and audit trails.
Key Contributions
- Architected the end-to-end RAG pipeline: document ingestion, embedding generation (OpenAI text-embedding-3-small), pgvector similarity search, and LLM response synthesis
- Built custom OAuth2 server handling authentication and authorization across platform services
- Designed distributed job queue (BullMQ + Redis) for reliable AI agent task orchestration with retries and dead-letter queues
- Implemented agentic execution layer: agents autonomously open PRs, run tests, and iterate on GitHub repositories
- Created the KVA design system and public website in Next.js 16