Praveen Kehelella

At Omnigate AI AB, I led development of an AI-powered IT support system for Region Västerbotten’s healthcare infrastructure. I built a RAG pipeline over Swedish-language Cosmic EHR documentation and shipped the platform to the region’s production servers.

Independently, I designed and deployed a production research knowledge platform that turns PDFs, DOIs, and YouTube videos into a searchable, evidence-backed knowledge graph. It combines PostgreSQL for structured metadata, Qdrant for hybrid semantic search, and an Obsidian vault for a human-readable concept graph, all served through FastAPI, React, and Celery workers. Answers are constrained to the ingested corpus and cite the exact page or video timestamp, and every graph relationship traces back to the source text chunks that support it. It runs on Docker Compose behind Caddy with invite-only access and automated GitHub Actions deploys.

At Rabot Inc., a venture-backed Vision AI startup for warehouse packing operations, I worked as a Robotics Software Engineer from 2021 to 2025, building the AI and edge layer behind overhead cameras that monitor pack stations in fulfillment centers. I developed a Python rule-based engine that turned YOLO object-detection outputs into human-friendly insights for automated content analysis, a product that directly helped the company win new customers. I used Google MediaPipe joint detection to run ergonomic analysis of employees at packing stations, and built a human-in-the-loop review system to monitor the health of AI features running on production devices. On the ML-Ops side, I created Docker containers for training deep learning object-detection models. I also supported the fleet of edge devices running AI on-device, helping scale from a handful of units to hundreds across the US in two years.