Open to opportunities

Pawan PAHUNE

@pawanpahune

I build production-ready agentic AI, RAG, and machine learning systems for financial, compliance, and scientific workflows.

India

Message

What I'm looking for

I'm looking to build production-grade AI, agentic RAG, multimodal LLMOps, and machine learning systems where I can turn research into measurable automation and real-world outcomes.

I've built agentic AI and RAG systems for financial intelligence, compliance auditing, and algorithmic trading. At Courser Park Advisors LLP, I delivered a multi-agent financial RAG architecture covering 700+ NSE stocks, reducing manual workflow time by at least 40%.

At Alliance of Bioversity International & CIAT, I engineered an AI-driven soil metagenomics pipeline and crop classifier across 2,944+ taxa profiles and 200+ samples, reaching 93.2% stratified five-fold cross-validation accuracy. I also developed quantum machine learning pipelines for highly imbalanced supply-chain backorder prediction with the Quantum Research Group.

My projects span LangGraph, hybrid retrieval, multimodal LLMOps, deep learning deployment, custom VLMs, and Stable Diffusion. I won Airothon 2025 with FINBUZZ.AI, a production-ready agentic finance platform combining MCP tools, financial RAG, and trading automation.

Experience

Work history, roles, and key accomplishments

Data Science Intern

Alliance of Bioversity International and CIAT

May 2026 - Jul 2026 (2 months)

Engineered an end-to-end AI-driven soil metagenomics pipeline to build an ML-based crop classifier on 2,944+ taxa profiles and 200+ samples. Preprocessed taxa with CLR transform, trained XGBoost classifier, benchmarked 10+ models achieving 95.5% train accuracy and 93.2% CV accuracy, and used SHAP for explainability.

Python Machine Learning XGBoost MLFlow DVC SHAP data preprocessing CLR Transformation Classification

Project Intern

Quantum Research Group

Apr 2025 - Aug 2025 (4 months)

Developed hardware-efficient Quantum Machine Learning models via Qiskit to optimize highly imbalanced data for supply chain prediction. Designed a 3-qubit supply-chain backorder prediction pipeline using hybrid QNN and VQC models on 14K samples with 1:200 class imbalance, achieving 85% stratified CV accuracy.

Quantum Machine Learning Qiskit XGBoost CatBoost LightGBM Imbalance Data

Agentic AI Intern

Courser Park Advisors LLP

May 2025 - Jul 2025 (2 months)

Delivered an end-to-end multi-agent financial RAG architecture using LangGraph to automate comprehensive analysis for 700+ NSE stocks. Built 10+ tools integrating MCP, established Hybrid RAG pipelines with Vector DBs, Knowledge Graphs, BM25+FAISS and ColBERT reranking, and validated with LangSmith and RAGAS.