At InnovationM, I built backend services for a production LLM interviewing platform, including agentic workflows and a hybrid evaluation engine that reduced LLM calls and inference cost per interview. I also owned incident response for an agentic-loop failure and shipped safeguards that reduced generated questions per turn from 60 to 1.
At Datopic Technologies, I deployed FastAPI OCR and LLM services and tuned inference to cut output latency by 50–60%. I also built RAG pipelines and document-intelligence services, and fine-tuned LLMs with LoRA/QLoRA.
