At Omilos, I architected document-search RAG pipelines using pgvector to retrieve information across 17M+ records, and optimized a production LLM to run 3x faster while maintaining response quality. I also evaluate model performance and iterate on prompting and retrieval configurations.
I built Nayamithra AI by fine-tuning Qwen with LoRA and QLoRA on 2TB of Indian legal data, then deployed it as an AWS production API serving 10,000+ users. I also created CORTEX, an open-source agentic coding assistant published to npm, and built ClinicalTrialMatchEnv, a deployed reinforcement learning environment for matching cancer patients to clinical trials.
