utpal@portfolio

> whoami▌

Utpal Das

GenAI Engineering Manager

RAG & Knowledge Graph Architect

Multi-Agent Systems Engineer

10+ Year Python Developer

> tech▌

Python FastAPI Databricks LangChain LangGraph Neo4j Docker
View Projects Download Résumé Contact

~/about.md

About

10+ years driving complex, cross-functional technical programs from ambiguous scope to production launch. Led delivery of four GenAI platforms for Fortune 500 clients (Pfizer, Allianz), coordinating engineering and client stakeholders across ingestion, retrieval, evaluation, and deployment.

Comfortable going deep on architecture — RAG, knowledge graphs, multi-agent orchestration — while owning program risk and launch readiness. Still writes code every night; the projects below are the proof.

~/experience.log

Experience

  1. Mar 2026 — Present Deloitte India · Engineering Manager, Technical Program & Architecture Lead Client: Allianz

    Agentic Data Migration Platform

    • Architected and led delivery of a 6-agent Claude Code system — clone, repo analysis, dataset lookup, code rewrite, output verification, feature-branch push — automating Foundry-to-Databricks migration end to end.
    • Cut delivery time up to 90% and migration cost up to 95% versus manual migration, while raising output per engineer.
    • Reduced per-run token cost with a terse-prompting ("Caveman") strategy and shipped an end-of-run report tracking assets migrated and tokens spent.
    • Integrated the pipeline across five external systems via MCP: Foundry, Databricks, Atlassian, GitMCP, Synapse.
  2. Jul 2021 — Mar 2026 Deloitte India · Engineering Manager, GenAI Platform Lead Client: Pfizer

    Oncoscout · Mapii · Constellation

    • Oncoscout — knowledge-graph chat over Neo4j/Cypher with single- and multi-hop retrieval, synonym lookup, and multi-modal SharePoint ingestion. Owned engineering execution and resolved a Cypher performance / hybrid-RAG stability issue that protected the launch timeline.
    • Mapii — shipped configurable chat and temperature settings for Pfizer Oncology, then led cross-team evaluation of clusterRAG/graphRAG as the next retrieval architecture.
    • Constellation — an enterprise RAG platform for Q&A, search, and summarization. A RAGAs-based evaluation gate cut hallucination-flagged responses by an estimated 30%. Negotiated production scope (SSO, audit logging) directly with client stakeholders. Ran on AWS EC2/Docker at an estimated 99.5%+ uptime, ~2,000+ queries/day.
  3. Jun 2019 — Jun 2021 Mindfire Solutions · Senior Python Developer Client: BrightEdge

    Spark (SEO platform)

    • Built crawler/parser infrastructure processing an estimated millions of pages a month across 50+ client domains to feed SEO ranking recommendations.
    • Shipped "Save for Later" bookmarking, adopted by an estimated 10,000+ users.
    • Architected a Google Maps → Leaflet.js migration saving an estimated $50K+ a year.
  4. Jul 2018 — May 2019 Cognizant · Python Developer Client: Walmart

    Partnerbot

    • Built an NLP/NLU seller-support bot (RASA) and a Flask template-generation service that standardized onboarding data (XML/Excel/JSON) for Walmart Marketplace sellers.
  5. Oct 2015 — May 2018 Infosys · Systems Engineer Client: Aetna

    PAR UI

    • Designed a MERN-stack policy/agent lookup application and the regression test framework that became the team's automated QA baseline.

~/skills.json

Skills

ai_llm

RAGGraphRAGHybrid RAG Knowledge GraphsNeo4jCypher LangChainLangGraphMulti-Agent Orchestration Claude CodeMCPLLM Evaluation (RAGAs)

platform

PythonFastAPIReact TypeScriptAWSDatabricks PostgresDocker

leadership

Technical Program ManagementSolution Architecture Cross-Functional DeliveryStakeholder Negotiation Risk & Launch Readiness

certifications

  • Google Certified GenAI Developer
  • Neo4j Certified Graph Data Science Developer
  • HackerRank Certified Python, Java, SQL Developer
  • Deloitte Certified Project Manager

education

  • B.Tech, Electrical & Electronics Engineering
    Siksha 'O' Anusandhan University · 2011 – 2015

~/projects/

Projects

enterprise client work · code under NDA

agentic-data-migration

NDA

Allianz

Six orchestrated Claude Code sub-agents automate a full Foundry-to-Databricks migration.

Claude CodeMCPDatabricksFoundry
Details

Up to 90% faster, up to 95% cheaper than doing it by hand — with a terse-prompting ("Caveman") strategy to keep per-run token cost down.

oncoscout

NDA

Pfizer Oncology

Knowledge-graph chat with single- and multi-hop retrieval, synonym lookup, and multi-modal document ingestion.

Neo4jCypherHybrid RAG
Details

Owned engineering execution and resolved a Cypher performance / hybrid-RAG stability issue that protected the launch timeline.

constellation

NDA

Pfizer

Enterprise RAG platform for Q&A, search, and summarization.

FastAPILangChainAWSDocker
Details

RAGAs eval gate cut hallucination-flagged responses ~30%. ~2,000+ queries/day at ~99.5%+ uptime on AWS EC2/Docker.

mapii

NDA

Pfizer Oncology

Configurable chat and temperature controls for Pfizer Oncology.

LangChainGraphRAGClusterRAG
Details

Led the cross-team evaluation that got clusterRAG/graphRAG adopted as the program's next retrieval architecture.

personal nights & weekends · code is public

agent-ops

Live

Jira for AI agent fleets: natural-language briefs become tasks, a routing matrix assigns the model by criticality × size.

Node.jsClaude CodeLLM routing
Live Demo GitHub

rock-paper-scissor

Live

Real-time 3D multiplayer, built on Three.js with zero build step. Procedural geometry, synced countdown.

Three.jsWebSocketsWeb Audio
Live Demo GitHub

ascii_image

Live

Drop an image in, get ASCII art out — entirely in the browser, no server round-trip.

JavaScriptImage processing
Live Demo GitHub

pyStocks+

Public

A BUY/SELL/HOLD signal derived from live NSE promoter-shareholding trends.

PythonFlaskNSE data
GitHub

~/ai-workspace/

AI Workspace

Current focus — what's actually in production or in daily use, not a buzzword list.

RAG Systems
Multi-Agent Orchestration
MCP (Model Context Protocol)
Claude Code
Knowledge Graphs (Neo4j)
LangGraph
Vector Retrieval (FAISS)

~/blogs/

Blogs

Nothing published yet — check back soon.

In the meantime, here's my YouTube channel.

~/contact.sh

Contact

$ contact

Email: utpaldas.6osm@gmail.com

LinkedIn: linkedin.com/in/utpal-das

GitHub: github.com/UtpalDas6

> Send Message