Debasish Maji — AI Engineer and Instructor

Your Instructor

Debasish Maji

Senior AI Engineer who built Rovo Agent at Atlassian and scaled ML systems at PhonePe (550M+ users). Now teaching working professionals to build production AI systems through live weekend bootcamps.

200+
Students taught live
120
Live sessions per batch
20
Industry projects built together
8+
Years in production AI

Production AI Experience

Everything taught in the bootcamp comes from shipping real AI systems at scale.

Atlassian

·Senior AI Engineer — Rovo Agent2022–2025

Built Rovo Agent — Atlassian's AI assistant that searches across Jira, Confluence, Slack, and Google Drive to answer employee questions. Designed the multi-tool agent architecture, RAG pipelines, and the ReAct execution loop that powers autonomous multi-step reasoning.

  • Architected the agent loop: reason → tool selection → execution → observation → iterate
  • Built RAG pipelines across 5 enterprise data sources with hybrid search
  • Shipped to 200K+ Atlassian Cloud customers
  • Led evaluation framework: LLM-as-judge + human review for agent quality

PhonePe

·ML Engineer2019–2022

Built and scaled ML systems serving 550M+ users on India's largest payments platform. Transaction fraud detection, recommendation systems, and real-time feature engineering at massive scale.

  • Fraud detection model processing 100M+ daily transactions
  • Recommendation engine driving 15% increase in merchant discovery
  • Real-time ML pipeline with sub-100ms latency requirements
  • Scaled from 10M to 550M users while maintaining model accuracy

Why This Matters for Your Learning

Real architecture, not theory

Every pattern taught — ReAct loops, tool selection, RAG chunking — comes from production systems serving real users, not textbook examples.

Scale-tested approaches

The ML patterns you learn were validated at 550M user scale at PhonePe. You learn what works under load, not just in notebooks.

Current industry knowledge

Direct experience with LangChain, Claude, GPT-4, vector databases, and multi-agent systems as they're used in production today.

Honest failure stories

You hear what went wrong — failed RAG experiments, hallucination incidents, cost overruns — so you can avoid the same mistakes.

Learn directly from someone who builds AI agents in production

The next cohort includes 120 live sessions across 20 weekends. Limited to 200 seats for meaningful instructor interaction.

View the Bootcamp