Headline

Data Scientist @Tata Technologies

About

Data Scientist with hands-on experience building production-grade generative AI systems, including agentic chatbots using LangGraph and ReAct-style reasoning, RAG pipelines with LangChain, hybrid retrieval, and Hugging Face embeddings. Skilled in evaluating LLM applications using RAGAS, DeepEval, and LLM-as-a-Judge techniques, deploying scalable services with FastAPI and Celery, and fine-tuning deep learning models with PyTorch. Experienced across the ML lifecycle — from model development and experimentation to production monitoring with MLflow — with a strong foundation in Python, cloud-based GenAI services (Azure), and secure, system design.

Location

Mumbai

AI stack

antigravity

claude code

hugging face

langchain

langgraph

sora

Work experience

AI Engineer at Tata Technologies, Aug 2025 - Present · 1 y 2 mo, Mumbai, Maharashtra, India, Built a production-grade Agentic AI chatbot based on ReAct framework using LangGraph for domain specialized text summarization and tool-calling, leveraging Azure GenAI Services (Azure OpenAI LLMs) and FastAPI, with iterative prompt design and refinement to improve output quality and relevance. Designed and implemented a RAG pipeline via LangChain — including semantic chunking of knowledge sources, Hugging Face embeddings, and hybrid retrieval (vector + keyword search) to retrieve from a ChromaDB vector store. Reduced overall latency in chatbot by over 40% by incorporating semantic caching and using websockets. Engineered a secure chatbot using LangChain and a locally deployed Ollama LLM model to ensure strict data confidentiality, integrating PostgreSQL and pgvector for cross-application similarity search with contextual compression to manage context window constraints, while enforcing dynamic role-based access control (RBAC) to prevent unauthorized data retrieval. Built custom evaluation harnesses using LLM-as-a-Judge techniques alongside RAGAS and DeepEval, continuously tracking live performance and tool-calling correctness, troubleshooting issues and refining models to ensure high reliability. Built a XGBoost classification model to predict churn for a commercial vehicle monitoring platform, analyzing large unstructured customer usage datasets. Built the complete modeling pipeline, with optimized hyperparameter tuning, improving customer churn detection accuracy by 12%. Presented model findings using SHAP for interpretability and estimated revenue retained through targeted model-driven outreach. Performed anomaly detection for a platform using Isolation forest to identify high-risk customer segments and abnormal patterns, achieving a Precision@K of 70%. Acted as a thought partner to business units by delivering original analysis that enabled targeted, data-driven customer support. Deployed, monitored, and maintained ML models and chatbots in production on AWS using Docker for containerized deployment, leveraging MLflow for experiment tracking, model versioning, and drift detection. Built and maintained CI/CD pipelines via Github actions to automate testing and deployment of ML models into existing pipelines, enabling scalable and reliable delivery.

AI Engineer at Tata Technologies, Aug 2024 - Aug 2025 · 1 y, Pune District, Maharashtra, India

R&D Intern at Tata Technologies, Feb 2024 - Jun 2024 · 5 mos, Pune/Pimpri-Chinchwad Area, Developed Python scripts using pandas and numpy to automate identification of bolted joints prone to failure in an automotive tow kit assembly reducing upto 3 hours of manual identification Created Tableau dashboard for better visualization of failed bolted joints.

Education

Bachelor of Technology - BTech, Mechanical Engineering, Vellore Institute of Technology

Public links

github: https://github.com/arnavxx7