Headline

Senior Data Scientist | Claude Certified Architect | GenAI & LLM Engineer | LangChain · RAG · Multi-Agent Systems · MCP | Turning LLMs into Production Infrastructure

About

I build GenAI systems that survive contact with production — not just demos that work in a notebook. Over 6+ years in ML, NLP, and Generative AI, I've shipped systems that actually carry business weight: a multi-agent RAG chatbot (LangChain + LangGraph) automating 80% of customer queries at 92% task-completion accuracy, an XGBoost pharmacist-alert model processing 800 cases/hour at 8x the target throughput, and a LightGBM credit-risk engine that cut loan defaults by 18% inside a regulated banking workflow. What I actually do: → Design and ship multi-agent RAG pipelines (LangChain, LangGraph, MCP) for enterprise chat and search → Fine-tune LLMs on domain data (Lamini) and lock down output reliability with RAGAS eval + Pydantic enforcement — cut hallucinations 31%, schema errors 87% → Build semantic search at scale — FAISS/Pinecone + MiniLM embeddings across 50K+ documents, sub-second retrieval → Ship classical ML that regulators can actually audit — SHAP-explained XGBoost/LightGBM models in banking and healthcare → Deploy all of it properly — FastAPI, Docker, Kubernetes, MLflow, CI/CD, not just Jupyter notebooks Right now at Fractal Analytics, I lead GenBI evaluation work — benchmarking Databricks Genie, Snowflake Cortex, and Power BI Copilot against a 36-prompt scoring library so business teams pick tools on evidence, not vibes. Before that, six years at TCS building production ML for banking and healthcare clients (Optum, and others) where "the model works" wasn't enough — it had to be explainable, monitored, and compliant. I'm a Claude Certified Architect (Anthropic) and hold the Microsoft Power BI PL-300 certification. Always happy to talk RAG architecture, agentic system design, or where GenAI hype meets production reality — reach out or drop a message. Skills: Python · LangChain · LangGraph · RAG · RAGAS · LLM Fine-Tuning · Prompt Engineering · MCP · FAISS · Pinecone · LightGBM · XGBoost · SHAP · FastAPI · Docker · Kubernetes · MLflow · AWS SageMaker · Azure ML · Power BI

Location

Mumbai Metropolitan Region

AI stack

claude code

hugging face

langchain

langfuse

langgraph

mcp

Work experience

Senior Data Scientist at Fractal, Apr 2025 - Present · 1 y 6 mo, Mumbai, Design and deploy production GenAI and BI-intelligence systems for enterprise banking and healthcare clients, from multi-agent RAG chatbots to cross-platform GenBI benchmarking. ->Led production of a multi-agent RAG chatbot (LangChain + LangGraph), automating 80% of customer queries at 92% task-completion accuracy across web and mobile ->Fine-tuned LLMs with Lamini on 10K+ domain documents, boosting answer precision 38% while maintaining full regulatory compliance for a banking client ->Cut hallucination rates 31% and API schema errors 87% by integrating RAGAS evaluation and Pydantic output enforcement into a reusable LLMOps framework ->Engineered a multi-engine GenBI evaluation harness (Databricks Genie, Snowflake Cortex, Power BI Copilot) with a 36-prompt scoring library, replacing anecdotal tool selection with quantitative decision intelligence

Senior ML Engineer at Tata Consultancy Services, Apr 2021 - Apr 2025 · 4 y, On-site, Led ML/AI initiatives for healthcare and banking clients, shipping models from prototype through regulated, monitored production pipelines. ->Built an XGBoost rare-event classifier on a 0.74% event-rate prescription dataset, achieving 76% recall and deploying the first AI-driven alert in a pharmacist verification pipeline ->Scaled inference to 800 cases/hour — 8x the 100/hour business target — while holding failure rate below 0.3% ->Delivered SHAP explainability on prescription risk drivers, making model decisions auditable for clinical and compliance teams ->Engineered a version-lineage data pipeline linking raw and pharmacist-verified prescriptions, eliminating mismatched training comparisons

IT Analyst / Python Developer at Tata Consultancy Services, Dec 2018 - Mar 2021 · 2 yrs 4 mos, Hybrid, Built the ML-based credit eligibility engine that became a core input to bank loan decisioning. ->Built a LightGBM + SMOTE credit scoring model achieving 89% AUC-ROC and 82% recall, reducing loan defaults by 18% ->Compressed 50–64 raw features by 60% through feature engineering while improving processing efficiency by 31% ->Applied SHAP to produce regulator-compliant, auditable explanations for loan officers ->Added post-COVID drift monitoring to sustain model reliability through market volatility

Internship at Mahagenco, Dec 2016 - Jun 2017 · 7 mos, Uran

Engineering mechanic and internal combustion engine at Multicon, Feb 2015 - Jun 2015 · 5 mos, Mumbai Area, India

Education

Bachelor of Engineering, Mechanical Engineering, Thakur College of Engineering and Technology

btech in mech, Mechanical Engineering, Thakur College of Engineering & Technology Shaymnarayan Thakur Marg Thakur Villaige Samata Naga

Public links

github: https://github.com/ankur19961

ankur19961-portfolio-app-app-x8yyio.streamlit.app: https://ankur19961-portfolio-app-app-x8yyio.streamlit.app/