AI/ML Engineer - FUTRAN SOLUTIONS - Pune, India
(2026-07)
Project: Enterprise Insurance Intelligence Platform
- Working on a production-grade Retrieval-Augmented Generation (RAG) platform using LangChain, Claude 5, AWS Lambda, and Amazon S3 for intelligent insurance policy analysis and AI-powered report generation.
- Contributing to an event-driven architecture integrating .NET, Amazon S3, and AWS Lambda for automated document processing and retrieval workflows.
- Developing document-grounded AI workflows that enrich user queries with policy context to generate detailed reports containing coverage insights, policy metadata, and contextual analysis.
AI/ML Engineer - IMPROZO - Pune, India
(2025-07 - 2026-05)
- Built a production Conversational AI Agent on Microsoft Copilot Studio, enabling 200 medical sales reps to debrief HCP meetings via speech on M365 mobile app, automatically extracting 12 structured fields from natural speech and eliminating manual notepad-to-CRM data entry entirely
- Architected intelligent ID resolution pipeline, extracting field values from speech using LLM prompting and mapping them to corresponding IDs across Snowflake and Microsoft Dataverse tables to populate TIKA CRM automatically, while displaying human-friendly field names to reps
- Reduced post-meeting CRM documentation from 1 hour to under 2 minutes per rep across 200 medical sales reps, saving hours of manual data entry daily and eliminating ID lookup errors entirely
- Architected a real-time document compliance pipeline on Snowflake Cortex, automating 5 domain-specific compliance checks across 2,000+ pharmaceutical documents via SharePoint to Snowflake integration using Workato
- Engineered detailed domain-specific prompts per compliance check with cross-document consistency validation, delivering 1-paragraph AI summary and structured pass/fail results to Power Apps dashboard in under 2 minutes
- Mitigated LLM hallucination risk through document grounding, chain-of-thought citation prompting, structured JSON output enforcement, and human-review flagging — reducing compliance review from 2 days to 2 minutes
