GenAI Engineer Intern
Edelweiss Life Insurance
May 2025 - Aug 2025 | Mumbai, India
GenAI-powered Business Insights and Visualization
- Built a serverless RAG app on AWS Bedrock (Claude 3.5 Sonnet) enabling real-time conversational access to AWS Data Lake, replacing 200+ static BI dashboards and enabling Rs. 300,000+ in estimated annual operational cost savings.
- Engineered Lambda functions for query validation, Redshift schema extraction, OpenSearch retrieval, and LLM inference; achieved 99% API uptime and reduced 8,700+ annual emails per user via automation.
- Boosted SQL generation accuracy by 28% using Cohere embeddings + OpenSearch retrieval for few-shot prompting. Integrated API Gateway + Streamlit UI, serving 1.5K+ daily active users with real-time tables/visuals.
- Fine-tuned CodeLlama-7B-Instruct using QLoRA and deployed it on NVIDIA TensorRT-LLM optimized SageMaker endpoints, accelerating inference by 1.8x and reduce memory usage by 50%.
Data Science Project - Policy Grievance and Mis-selling Prediction
- Productionized a top-5 LightGBM + CATBoost ensemble to flag grievance/mis-selling risk across 809K+ policies, surfacing 97% of grievance cases in the top decile, empowering agents to preempt complaints.
- Developed a scalable ML pipeline on AWS SageMaker, containerized with ECR, and orchestrated automated weekly predictions reports with AWS Lambda and AWS EventBridge Scheduler cron jobs, reducing prediction latency by 3x.
- Operationalized model outputs by tagging high-risk users as “Handle with Care” in internal systems and routing them to trained agents in real time, driving a 23.8% drop in 3-month rolling average of complaints.
RAGLLMNVIDIA TensorRT-LLMClaude 3.5 SonnetLlamaPrompt EngineeringAWSLambdaSageMakerBedrockHuggingFaceCohereOpenSearchStreamlitAPI GatewayPythonBashDockerLightGBMCatBoostEventBridgeECR