Varish Kapur

I’m a Backend AI Engineer building RAG-based retrieval pipelines and LLM-powered backend systems for production use at FlyRank AI, including scalable inference infrastructure and production-grade APIs.

In my projects, I’ve combined classical ML, deep learning, and reinforcement learning into deployable products—like FinSentinel’s hybrid anomaly detection with a DQN agent and an LLM layer that generates analyst-style reports (under review at an Elsevier journal).

I also built FULLHOUSE, a clinical decision-support system over rare-disease data, using UMAP + HDBSCAN clustering with XGBoost, then integrated Groq-hosted Llama 3.3 to produce differential-diagnosis explanations for clinicians.

On Meeting Intel, I shipped an end-to-end meeting-intelligence pipeline using Groq Whisper Large V3 for fast transcription and Llama 3.3 for structured extraction, deployed as a FastAPI service.