• Designed and implemented a RAG retrieval service using Amazon OpenSearch, Bedrock Knowledge Bases

and S3, with fine-grained access control.

• Implemented query decomposition improving contextual relevancy by 7% and answer relevancy by 10%.

• Developed LLM guardrails for risky, malicious and out-of-domain detection, reducing unsafe responses.

• Created automated evaluation platform (using DeepEval, Ragas) to benchmark and improve RAG quality.

• Built RAG applications using GPT-4V’s multimodal capabilities (images, video, text), achieving 80%+ retrieval accuracy across 300+ documents, videos and images.

• Developed a text-to-SQL RAG framework with OpenAI, converting natural language into SQL for advanced data access.

• Implemented Agentic AI using LangChain’s ReAct framework, integrating 20+ internal/external APIs via function calling for advanced automation.

• Designed and optimized prompt templates for varied emotional tones and user scenarios.

• Improved user experience by adding FAQs, prompting type-ahead questions and follow-up questions.

• Led a team of 5 to develop a Python app to process video streams, detect damage with AI and generate condition reports.

• Built REST APIs in Java for batch PDF generation using AWS SDK, CI/CD pipeline and monitoring dashboard.

• Led seismic data processing projects under tight deadlines and managed end-to-end data workflows for petabyte-scale datasets, including cleaning, modeling, validation and visualization.

• Developed and deployed deep learning models for image processing and geological feature identification.