Vinay Shirole

About Vinay Shirole

Data Scientist and AI/NLP Engineer with strong experience building scalable data platforms,machine learning systems, and agentic AI solutions in production environments.I am currently working as a Data Scientist (Contract) at Cummins, where I contribute toadvanced data and AI initiatives across cloud and enterprise platforms.Most recently, I am working on a POC (proof of concept) focused on building a multi-agentarchitecture across platforms such as Databricks, Snowflake, and Palantir. The systemenables intelligent agents to communicate with each other via MCP (or similar orchestrationmechanisms) to solve complex, cross-platform data and analytics tasks. This work focuses on building enterprise-ready, governed agentic systems rather thanstandalone LLM demos.On the Databricks side, I have implemented a multi-agent setup with:• A document-based RAG model for contextual retrieval and reasoning • A Genie space as an interactive analytical tool • An orchestration layer where the agent dynamically selects the appropriate tool based on user intent and query complexity Beyond agentic AI, my core strengths include:• Designing and optimizing ETL/ELT pipelines using PySpark, SQL, Databricks, and Delta Lake • Building and deploying ML & Deep Learning models (CNNs, NLP, GenAI, LLM fine-tuning) • Managing the ML lifecycle using MLflow, APIs, and CI/CD pipelines • Working across Azure, AWS, and GCP-based data ecosystems • Translating business and research problems into scalable, data-driven solutions Previous experience includes:• Large-scale data ingestion, validation, and CI/CD automation at Cummins • Geospatial ML and satellite imagery analysis for flood risk prediction at Indiana University • Enterprise data migration and pipeline orchestration at Capgemini I am open to opportunities in:Data Engineering | Data Science | Machine Learning | GenAI / Agentic AI | Analytics Engineering Open to US & Europe | Remote, Hybrid, or On-site

Experience

1. Data Scientist

   Cummins Inc.

   Jun 2025 — Present · Columbus, IN, US

   Modernized legacy Fortran systems by leading migration to Databricks, implementing parallel processing with Ray, and enabling MLOps adoption across engineering teams.• Architected a multi-agent Generative AI system using LangGraph for orchestration and MCP protocol for tool integration; developed an LLM-driven agent (Claude Sonnet 3.7) capable of autonomously selecting from 5+ tools, including RAG (Unity Catalog Vector Search), SQL generation (Genie Spaces), and web search based on natural language intent.• Integrated Microsoft Copilot Studio with Databricks-managed MCP servers via OAuth client configuration, enabling secure LLM access to enterprise data through the Model Context Protocol (MCP).• Developed a RAG retrieval system leveraging Databricks Unity Catalog Vector Search and databricks-embedding-large; reduced hallucination rate from ~30% to <5% by transitioning from fixed token chunking (512 tokens) to recursive chunking and refining system prompts.• Optimized GPT-4 few-shot prompting for a rule translation system that converts natural language pricing rules into executable Python constraints; reduced manual effort by 95%(40 hrs → 2 hrs) using structured JSON outputs validated with Pydantic, and ensured accuracy within ±5% MAE across 6,500+ product configurations.• Migrated pricing analytics pipeline from Pandas to PySpark, reducing execution time by 40%(10 hrs → 6 hrs) and lowering compute costs through parallelized processing of 6,500+ product combinations.

Education

• Indiana University Bloomington

  Master of Science - MS, Data Science

• St Marys Junior College, Vashi

  HSC, Computer Science

• SBOA Public School - India

  SSC

• University of Mumbai

  Bachelor of Engineering, Computer Engineering