Pranav K R

I'm building agentic AI systems at Auvoya that turn travel data into grounded recommendations. I developed pipelines across 250+ destinations, bringing accommodations, transport, weather, blogs, Reddit discussions, and destination metadata into Neo4j and Qdrant.

I fine-tune and apply BERT-based NLP models to transform large-scale Reddit discussions into structured signals for retrieval and recommendation workflows. I also built a reusable MCP server for source discovery, content extraction, Qdrant ingestion, and Neo4j-ready artifacts.

I designed a DeepAgents-based orchestration layer and a GraphRAG retrieval pipeline using Neo4j traversal, Qdrant vector search, hybrid retrieval, and deduplication. I evaluate recommendation quality through context precision, recall, faithfulness, hallucination detection, and answer relevance.

My computer vision projects include a YOLOv11 and StrongSORT football tracking system with 88% mAP@0.5, plus a YOLOv8 dental anomaly detector trained on 7,800+ X-ray images and exported to TFLite for edge deployment.