GenAI & NLP systems for GxP-regulated pharmaceutical environments
AI/ML Engineer with 11 years at Tata Consultancy Services (9+ years active delivery, plus a 2-year sabbatical for a full-time MBA), currently working on a major UK pharmaceutical account. Transitioned from Business Analysis and Scrum Master leadership — including leading a 26-person team on a £13M digital transformation programme — into a hands-on individual-contributor AI/ML role. Builds multi-agent architectures, NLP/LLM pipelines (RAG, fine-tuning, evaluation) and classical ML models, with a track record of translating regulated-industry (GxP/pharma) business problems into deployed technical solutions. Combines strong stakeholder and delivery leadership with applied machine learning, deep learning and LLM engineering skills.
| Project | What it does | Stack |
|---|---|---|
| agent-alpha 🚧 | Multimodal agentic field-service copilot — detects equipment faults from images (YOLOv8), diagnoses root cause, retrieves relevant SOPs via RAG, and generates step-by-step repair plans through a 5-agent pipeline (Vision → Diagnostics → Retrieval → Planning → Supervisor) with a full audit trail. | Python, YOLOv8, LangChain, FAISS, Streamlit, FastAPI |
| cortex-sre 🏆 | Autonomous SRE agent built on Coral SQL — joins Sentry, GitHub, and Slack signals in one query, then closes the loop by patching code locally (Ollama), opening a real GitHub PR, and notifying Slack. Built for the Coral Bean hackathon. | Python, FastAPI, Coral SQL, Ollama, Docker |
| Medical_Assistant_rag 🚀 | Production-grade, containerized Medical Assistant RAG app — indexes and queries medical documentation using a local quantized Llama-3.2-3B model, FastAPI, LangChain, and ChromaDB. Deployed to Hugging Face Spaces. | Python, FastAPI, LangChain, ChromaDB, Docker, llama.cpp |
| c5tree 📦 | A pure-Python, scikit-learn-compatible implementation of Quinlan's C5.0 decision tree — gain-ratio splitting, native missing-value handling, multi-way categorical splits, pessimistic error pruning. Published on PyPI with CI. | Python, scikit-learn API, GitHub Actions |
| VATSA 📝 | Video, Audio, Text, Sensory, Action — a concept paper and research roadmap for a unified five-modality AI architecture for safety-critical robotics, integrating vision, audio, text goal parsing, and spatial sensor fusion. | Research / Concept Paper |
🚧 = active build · 🏆 = hackathon project · 🚀 = deployed · 📦 = published package · 📝 = independent research
For the full story behind these — including career timeline, GxP compliance background, and doctoral research focus — see my interactive portfolio site.
- Languages & Frameworks: Python, SQL, FastAPI, LangChain, LangGraph
- Machine Learning: Regression, Decision Trees, Clustering, Bagging & Boosting, Hyperparameter Tuning, Model Interpretability
- Deep Learning & Computer Vision: Neural Networks, CNNs, VGG-16, Data Augmentation, Pytorch, TensorFlow, Keras
- NLP & Generative AI: LLMs, Retrieval-Augmented Generation, Prompt Engineering, LLM Fine-Tuning, Embeddings, Chunking, Retriever Architectures, LLM Observability & Evaluation (Weave)
- MLOps & Deployment: Docker, Azure Container Apps, Azure Document Intelligence, Azure OpenAI (GPT-4o), Model Deployment, API Development
- Data & Tools: NumPy, Pandas, Scikit-learn, Seaborn, EDA, ChromaDB, Hugging Face, Databricks
- Data Scientist Professional — DataCamp (2026)
- Building LLM Applications with Prompt Engineering — NVIDIA (2025)
- SQL Associate — DataCamp (2025)
- Professional Scrum Master I (PSM I) — Scrum.org (2023)
- Great Lakes ML Hackathon — 3rd Place (2026)
- Hackathon builder — Coral Bean (2026)
📫 Open to relocation (Bangalore) · Based in London, UK · +44 7424 696464