📌 Find me at
|
🚀 Senior Consultant – Data Science & AI @ EY | Microsoft Certified: Azure Data Scientist Associate AI & ML engineer with 10+ years of IT experience (7.5+ years hands-on in Data Science, Deep Learning & Generative AI) across Banking, Healthcare, Energy, and Tax. I specialize in LLMs, LangChain agents, RAG pipelines, and production-grade ML — leading enterprise GenAI initiatives such as EY's Sphere AI-enabled Tax Co-Pilot system, and shipping fine-tuned, guardrailed LLM applications end to end. Skilled across the full ML lifecycle — statistical modeling, deep learning & NLP, open-source LLM frameworks (Hugging Face, LangChain), and SOTA models (BERT, GPT, Llama) — from research and prototyping through Azure-based production deployment. Always happy to talk AI, ML, or anything tech-related! |
|
- 🧑💼 10+ years of IT experience (7.5+ years hands-on in Data Science & AI) across Banking, Healthcare, Energy, and Tax
- 🤖 Leading GenAI initiatives at EY — architected Sphere, an AI-enabled Tax Co-Pilot system, managing a team of 5 analysts building GenAI POCs with multi-agent orchestration (Chain of Thought & Tree of Thought prompting)
- 🔗 Built enterprise LangChain agent systems for compliance automation, tax-alert monitoring, code migration (SAS → PySpark, Postgres DBT → BigQuery DBT), and approval-engine automation — each cutting manual effort by 60–90%
- 🧠 Hands-on with Transformers, BERT, GPT, and Llama; RAG, fine-tuning, Hugging Face, and Vector Databases for reliable LLM applications
- 📊 Delivered predictive ML with measurable business impact — credit risk PD models (15% lower default rate), energy demand forecasting with LSTM (40% lower forecast error), delinquency prediction, and NLP-driven automation
- 🎓 M.Tech in Data Science & Engineering, BITS Pilani (CGPA 9.16/10)
| Company | Role | Duration |
|---|---|---|
| EY Global Delivery Services India LLP | Manager (Senior Consultant – Data Science & AI) | Jan 2022 – Present |
| Genpact India Pvt. Ltd | Assistant Manager | Mar 2021 – Jan 2022 |
| PwC India Acceleration Center | Senior Associate | Jul 2018 – Mar 2021 |
| TCS Analytics & Insights | Senior Business Analyst | May 2018 – Jun 2018 |
| Accenture Solutions | Analyst | Oct 2014 – May 2018 |
- Languages: Python, R, SQL
- LLMs & Generative AI: LangChain (agents & RAG), LangGraph, Hugging Face, Transformers, BERT, GPT, Llama, Prompt Engineering (CoT/ToT), Fine-tuning (QLoRA)
- RAG & Search Systems: Vector Databases, Azure AI Search, FAISS, Qdrant, Hybrid RAG, Graph RAG
- Machine Learning & Deep Learning: scikit-learn, XGBoost, PyTorch, TensorFlow/Keras, LSTM, Time Series Forecasting, NLP
- Cloud & MLOps: Azure, Azure MLOps, Azure AI Search, Docker, FastAPI, CI/CD
| Certification | Issuer |
|---|---|
| Microsoft Certified: Azure Data Scientist Associate | Microsoft |
| dataiku Certified: ML Practitioner | dataiku |
| IBM Data Science Professional Certificate | IBM (Coursera) |
| Deep Learning Specialization | deeplearning.ai |
| Natural Language Processing Specialization | deeplearning.ai |
| TensorFlow Developer Professional Certificate | Coursera |
| Business Transformation Through Machine Learning | IIM Calcutta |
| R Programming | Johns Hopkins University |
| Databases and SQL for Data Science | IBM |
| BCS ISEB Certified Business Analyst Professional (License No. 00325319) | BCS |
- Liver Disease Detection: Evaluation of Machine Learning Algorithms Performances With Optimal Thresholds — International Journal of Healthcare Information Systems and Informatics (IJHISI)
- Analysis of Machine Learning Algorithms for Email Classification Using NLP — IEEE International Conference on Computer, Electrical & Communication Engineering, Dec 2017
I write deep-dives on RAG, LLM fine-tuning, and agentic AI systems, published on Medium (via Towards AI):
| No. | Article | Published | Topics |
|---|---|---|---|
| 1 | Architecting State-of-the-Art Text-to-SQL Agents for Enterprise Complexity | Dec 14, 2025 | LLM, Machine Learning, GenAI |
| 2 | Building a Fully Self-Healing RAG System | Nov 30, 2025 | Deep Learning, LLM, RAG |
| 3 | Hybrid Graph RAG: Harnessing Graph and Vector for Financial Analysis | May 1, 2025 | Graph Database, NLP, Machine Learning |
| 4 | Building a Self-Improving RAG System with Reinforcement Learning | Apr 28, 2025 | NLP, LLMs, Reinforcement Learning |
| 5 | Fine-Tuning Phi-2 for Medical Reasoning with QLoRA | Apr 23, 2025 | Fine-Tuning, LLM, Healthcare |
| 6 | Decoding Money Laundering: LangGraph's Multi-Agent Blueprint | Mar 27, 2025 | Multi-Agent Systems, LangGraph, Data Science |
| 7 | Guardrails to Ensuring Safe and Relevant AI Interactions | Mar 24, 2025 | LangChain, Gen AI for Business, OpenAI |
📚 Read more on Medium
| No. | Repo | Description | Category |
|---|---|---|---|
| 1 | RL-Self-Improving-RAG ⭐ | RL-enhanced RAG system that continuously optimizes retrieval strategies for QA, achieving state-of-the-art accuracy via self-improvement. | RAG / RL |
| 2 | Langgraph_AML_Detection | Anti-Money Laundering detection powered by LangGraph's graph-based reasoning, with compliance reporting. | Agentic / FinTech |
| 3 | text-to-sql-drgc-agent | Production-ready Text-to-SQL using the DRGC (Decomposition-Retrieval-Generation-Correction) framework for complex enterprise queries. | NLP / Agents |
| 4 | Hybrid-Graph-RAG-Financial-Analyser | Combines vector retrieval with graph knowledge extraction for financial analysis of Apple's financial data. | Graph RAG / Finance |
| 5 | MedReason-LLM-QLoRA-Fine-Tuning | End-to-end fine-tuning of Phi-2 on the MedReason dataset using QLoRA for parameter-efficient clinical reasoning. | Fine-tuning / Healthcare |
| 6 | gemini-multimodal-rag | Document intelligence using Gemini AI to extract and reason across text, tables, and visuals. | Multi-modal RAG |
| 7 | Traffic-Flow-Prediction | Spatio-temporal traffic forecasting across 36 sensor locations in Northern Virginia / DC, 15 minutes ahead. | Time Series / Deep Learning |
| 8 | Chatbot-NeMo-Guardrails | Secure banking assistant on Azure OpenAI with content filtering, PII detection, and compliance controls. | LLM Safety / FinTech |
| 9 | Databrick-RAG | Production-ready RAG pipeline on Databricks + LangChain with optimized chunking, embedding, and retrieval. | Production RAG |
🔗 For deeper case studies, an interactive architecture diagram, and my writing — check out my full portfolio: subrata-samanta.github.io