TEJAS KAMBLE

Data Integrity Agent

• Building an AI-powered root cause analysis agent for a healthcare / life sciences data platform that investigates pharma client data questions over LAAD Rx, Mx, and Medicare Part D datasets using Lang-Graph, Fast-API, and Snowflake.
• Designed intake-to-investigation flow to classify client queries into hypothesis recipes, sync Salesforce case context and attachments, and conduct multi-step SQL analysis and report generation post human approval.
• Implemented recipe-driven investigation paths for volume and trend analysis, MPD eligibility and phasing, and financial field / OPC consistency checks per LAAD Investigation SOPs.
• Built the SQL planning, generation, execution, and retry loop with deterministic query construction, result analysis (CSV and charts), and structured RCA outputs persisted to Cosmos DB and Azure Blob.
• Developed evaluation cases from client investigation annotations and intake classification tests to assess conclusion quality and recipe routing accuracy.

One-Key VR Automation

• Developed a multi-agent Lang-Graph workflow for a healthcare data platform that validates healthcare professional and organization Verification Requests against One-Key DB and country-trusted sources (France, Italy).
• Designed Supervisor, Consistency, and Search and Summarize agents to classify request comments, compare OK DB matches, run external verification only when needed, and produce structured BDO decisions (create, update, invalidate, auto-validate, phone call).

Scenario Planning Tool

• Developed AI-powered scenario planning tool with interactive UI, streamlining project planning and enhancing accessibility for non-technical users.
• Optimized machine learning models for cost forecasting by integrating multiple data sources and validating results with GPD forecasting methods, enhancing predictive accuracy through feature engineering, cross-validation, and statistical techniques.

LLM Chatbot with RAG

• Built and deployed LLM chatbots with Retrieval-Augmented Generation (RAG) using Lang-Chain and Dataiku, enabling context-aware answers over internal knowledge bases.
• Improved response relevance through prompt engineering, dynamic retrieval tuning, and iterative evaluation of answer quality across user queries.

LLM Fine-Tuning

• Fine-tuned large language models using PEFT and LoRA techniques to improve model efficiency and responsiveness.