Nikhitha Reddy Vempalli
Applied ML, Agentic AI, RAG | Ex-S&P Global | 5x Hackathon Winner
Hyderabad
vnikhithareddy06@gmail.com
I build Applied ML and agentic systems that work in the wild. This summer at S&P Global, I engineered a multi-agent RAG assistant for commodity price forecasting that is now heading to production. Previously, I researched and built agentic LLM architectures at UNT. I secretly thrive on the unglamorous stuff: failure recovery, evaluation metrics that hold up, and making LLMs behave. Final-year Data Science undergrad at GNITS, 5x hackathon podium finisher, and open to ML/AI Engineering roles for 2027.
Summer Intern, Software Development (Data Science & Modelling)
S&P Global
- Built a multi-agent AI assistant with RAG (Bedrock, pgvector, FastAPI) for an internal commodity price-forecasting platform. It cut analyst data lookups from ~15 min to under 10 sec and is in QA for release.
Data Science Research Intern
University of North Texas
- Built spar.ai, an agentic LLM system (LangGraph, ChromaDB) that splits instructions into sub-prompts, generates test cases and debugs its own output. Reviewed 15+ papers on multi-agent LLM architectures.
Saga-MCP
A recovery layer for tool-using LLM agents. When a multi-step workflow fails midway, it rolls back completed steps in reverse order using human-approved compensating actions. On a controlled benchmark it kept state consistent at 0%, 25% and 50% injected faults.
BhashaMudra
A full-stack Indian Sign Language learning app with real-time gesture recognition, 3D avatars and adaptive quizzes. 99.96% accuracy on alphabets and numbers with a dual-model pipeline.
A prediabetes risk predictor for Indian adolescents that forecasts years to T2DM onset (R² 0.91), with SHAP explanations and auto-generated PDF reports.
B.Tech, Computer Science and Engineering (Data Science)
G. Narayanamma Institute of Technology and Science (for Women) · 9.31/10.0
Attendee · Oct 24, 2026