Shourya Singh

Built a system that automatically reads unstructured defence documents and extracts key people, places, and relationships, turning scattered text into a structured, searchable knowledge base (62% F1 accuracy across 5 entity and 7 relation types). Enabled analysts to ask questions in plain English and get grounded answers pulled directly from the knowledge graph using Neo4j and natural language to

Flagged $18M in recoverable spend by building a SageMaker + PySpark anomaly detection pipeline over 41K+ supplier contracts, surfacing 1,285 irregular agreements. Cut projected delivery losses by $14M annually by designing a GenAI validation pipeline on AWS Bedrock that corrected invalid vendor return addresses across 200+ countries.

Cut average debugging time from 4.2 to 1.7 hours per incident (800 engineering hours saved quarterly) by fine tuning a BERT model for a production log classification system on Azure at 91% accuracy.

Built a production WhatsApp AI agent for fitness businesses, automating live customer enquiries, qualification, and trial bookings end to end with zero human intervention. Engineered agent memory and per-user session state for multi-turn continuity, with structured JSON output routing decisions into booking, CRM, and payment tool calls.

Built an AI nutrition tracker where users log meals in plain English and get instant calorie and macro breakdowns. Matched free text meal descriptions to a database of 8,000+ foods using AI powered semantic search (RAG), returning accurate results in under 2 seconds.

Activities and societies: Vice President, Burning Ice student club, with 100+ members, Rs 4.2L raised, 8 events. Professional Muay Thai fighter, state powerlifting medallist.

Pursuing a B.Tech in Data Science and Engineering, expected to graduate in 2026.