Mayuna Gupta

About

Built and deployed an LLM-powered RAG system (PubChemRAG) for a 4D Mitospace project to compare mechanisms across a ~100-drug glossary and surface expected pathway/phenotypic differences in mitochondrial imaging. Worked closely with biochemistry and microscopy experts to design tiered evaluation benchmarks, iterating on prompts, retrieval quality (corpus hygiene, chunking strategy), and model outputs under GPU constraints using LangChain.

Experience

Education

Key Strengths

• Built and deployed PubChemRAG (LLM/RAG) to compare drug mechanisms and predict phenotypic changes in mitochondrial images
• Strong data quality focus: corpus hygiene, relevance filtering, and formatting normalization to improve retrieval quality
• Practical RAG tuning: adaptive chunk-size moderation and prioritizing abstracts/results for higher-signal context
• End-to-end LangChain implementation (loading, chunking, embedding, storage, retrieval, prompt templating, chaining)
• Structured evaluation design with domain experts using easy/medium/hard test tiers and LLM-call-based testing
• Makes model/retrieval choices based on task complexity and GPU/resource constraints; iteratively refines outputs with consistent formatting
• Effective collaboration with non-technical stakeholders (biologists/microscopy team) via iterative model versions and qualitative feedback loops