Vishal Kumar

About Vishal Kumar

I design and build production-grade AI systems — the kind that have to work reliably at scale, not just in demos.Over the past 9 years, I\'ve worked across banking, finance, automotive, and healthcare, moving from classical ML engineering into full Generative AI and agentic platform architecture. Today at Deloitte, I lead the design and delivery of systems that most teams are still figuring out how to prototype — multi-agent orchestration pipelines, RAG architectures with real retrieval and governance layers, LLM evaluation frameworks built on RAGAS and DeepEval, and agentic systems that connect directly to production ERP environments.What I care about is the work that happens after the model works — the fallback design, the observability layer, the token cost governance, the injection guardrails, the evaluation pipeline that catches regressions before users do. That\'s where production AI actually lives, and that\'s where I spend most of my time.Before Deloitte, I built ML systems for predictive maintenance at TCS, spend anomaly detection at American Express, and regulatory reporting pipelines at Citco — which gave me a foundation in the full ML lifecycle before Generative AI became the conversation.I\'m based in Pune and open to conversations about senior AI architecture, platform engineering, and agentic system design roles

Experience

1. Lead Data Scientist 2

   Deloitte

   May 2025 — Present · Pune, IN

   Lead architect and team lead for multiple internal AI products within Deloitte\'s Innovation & Technology team. I own the full system lifecycle across agentic pipeline design, RAG architecture, and LLM governance — managing a team of Data Scientists and Python Developers while driving technical direction across workstreams.Key systems I\'ve architected here include a production multi-agent defect resolution platform on LangGraph and LangChain with live ERP integration, a reusable RAG module with configurable vector store and model backends, an automated SOX control testing pipeline with OCR-based evidence capture, and an LLM evaluation framework built on RAGAS, DeepEval, MLflow, and Langfuse.

Education

• ITM UNIVERSITY, GWALIOR

  Bachelor's degree, Computer Science

  2012 — 2016

• Mithila Public School - India

  AISSE, Mathematics

  2008 — 2010

• Millia Convent English School, Purnea, Bihar

  AISSCE, Maths

  2010 — 2012

Skills

• English
• Microsoft Office
• Databases
• Powerpoint
• Leadership
• Windows
• Social Media
• C++
• Java
• Hadoop
• Management
• Microsoft Word
• Placement Test
• Teamwork
• Team Management
• Html
• C
• Programming
• Project Management
• Research
• Mysql
• Customer Service
• Cascading Style Sheets (Css)
• Customer Relationship Management (Crm)
• Public Speaking
• Sql
• Time Management
• Linux