Biruk Abza

Developed an AI-driven coding assistant for the MeTTa programming language using a RAG architecture.
Achieved 91% benchmark accuracy on a custom evaluation suite, significantly outperforming baseline models.
Reduced average response latency from 16s to 8s by implementing WebSocket streaming.
Designed a tool-augmented retrieval system for dynamic querying and code generation accuracy.
Deployed the AI assistant web platform on a DigitalOcean VPS, serving 200+ internal developers.
Built an MCP server integration for Claude Code.

Contributed to Rejuve.Bio's AI-driven platform for discovering anti-aging therapeutics.
Optimized genomic variant annotation workflows, cutting runtime by 66%.
Implemented real-time task feedback using WebSocket between Flask and Prefect services.
Engineered a literature mining engine using PubMed API and Sentence Transformers.