Open to opportunities

PRAVEEN V

@praveenv

Message

AI/ML engineer focused on agentic RAG systems, LLM evaluation, and physics-informed digital twins.

India

Message

What I'm looking for

I’m looking for an AI/ML role where I can build agentic LLM and RAG systems, ship reliable FastAPI/Streamlit services, and evaluate models rigorously—while applying ML to real systems like digital twins.

I build production-minded agentic AI for retrieval-augmented generation and benchmarking. In my Multi-Agent RAG project, I designed a 3-agent LangGraph pipeline (Research → Synthesis → Critique) with automated self-correction, reaching 0.95 Faithfulness and 0.97 Context Recall on RAGAS benchmarks.

I also focus on reliability, extensibility, and real-world behavior modeling. I created a multi-tenant auth and persistence layer with JWT, async SQLAlchemy/PostgreSQL, and Qdrant payload isolation to prevent cross-user access, and I reduced Groq API failures from ~50% to <5% using payload filtering and map-reduce summarization with adaptive retries. Beyond RAG, I built an agentic AI benchmarking system (8-node LangGraph) that reduced inference calls by 57× via batching and caching, and a hybrid physics–ML microgrid digital twin that achieved 0.9711 ROC-AUC (XGBoost) and 0.8646 R² (Random Forest) for blackout risk forecasting.

Experience

Work history, roles, and key accomplishments

FacetBench Agentic Benchmarking

Individual Project

Designed an 8-node LangGraph benchmarking system with modular scoring stages for agentic AI evaluation and deployed it with a frontend. Reduced inference calls via batching and caching, built a ChromaDB-based RAG scoring pipeline, and exposed the pipeline through a FastAPI REST API.

LangGraph LLM Evaluation Agentic AI Conditional Routing Tool Use Prompt Engineering fastAPI Streamlit ChromaDB Sentence Transformers Sentence Transformers (All MiniLM L6 V2)Ollama Batch Caching

Self-Correcting Multi-Agent RAG

Individual Project

Built a 3-agent LangGraph pipeline (Research → Synthesis → Critique) with automated self-correction and evaluated it on RAGAS benchmarks. Implemented multi-tenant JWT auth and persistence with per-user Qdrant isolation, improved cross-document retrieval reliability, and deployed a full-stack UI with evaluation dashboards.

LangGraph RAG Multi Agent Systems Prompt Engineering LLM Evaluation Ragas fastAPI React PostgreSQL SQLAlchemy Qdrant JWT GROQ LLaMA 3.3 70B Hugging Face Spaces Vercel Docker

Hybrid Physics–ML Digital Twin

Team Project

Built a physics-informed offline digital twin for an IEEE 33-bus microgrid using pandapower and scenario-based dataset generation for blackout prediction. Trained and evaluated hybrid ML models for blackout risk forecasting and created a Streamlit dashboard for simulation monitoring and predictive scoring.

Digital Twins PandaPower Newton Raphson BESS State of Charge Tracking Data Generation Machine Learning Supervised Learning Classification regression Feature Engineering XGBoost Random Forest scikit learn Pandas NumPy Matplotlib Streamlit Anomaly Detection Risk Scoring

Education

Degrees, certifications, and relevant coursework

Indian Institute of Information Technology, Nagpur

Bachelor of Technology (B.Tech), Computer Science and Engineering (AI/ML)

2023 - 2027

Activities and societies: Research Assistant: Self-Correcting Multi-Agent RAG (LangGraph) with automated critique; built multi-tenant auth/persistence + Qdrant isolation, reduced Groq failures (~50%→<5%), deployed on Hugging Face Spaces/Vercel. Projects: Agentic AI benchmarking system (LangGraph) and Hybrid Physics–ML digital twin for microgrid blackout prediction (pandapower; XGBoost/Random Forest) with Streamlit dashboard.

Pursuing a B.Tech in Computer Science and Engineering (AI/ML) at IIIT Nagpur (Aug 2023–May 2027). Coursework includes machine learning, artificial intelligence, data structures, databases, computer networks, and operating systems.