Suman MukherjeeData & AI Delivery Lead
Open to conversations Email

I take AI from POC to P&L.

An AI answers as me
Start as

Pick a question, or type your own.

An AI answers as me, grounded in my real work. If the live model is busy, you get a prepared answer instead, and the label says which. Please don't share personal details here.

Or jump toProgrammesLab filmsFit checkContact

Portrait of Suman Mukherjee, smiling, in a black T-shirt
Suman MukherjeeIndia · Global
Open to conversations

12+ years leading GenAI, ML, customer intelligence and cloud-data programmes across insurance, life sciences and retail. I move AI beyond demos into governed, audited production with clear ownership and measurable outcomes.

Now

Leading a ~$5M Data & AI portfolio in Gulf insurance

Proof in numbers

12+

Years in AI & data

$5M

Portfolio owned

~30%

Lead conversion lift

500TB+

Cloud data migrated

Selected programmes

I move AI beyond demos into governed, audited production with clear ownership and measurable outcomes.

Now
Illustration of stacked shields protecting groups of people

Insurance · Gulf · 2024–now

Enterprise AI Transformation

Delivery governance for a ~$5M Data & AI portfolio: propensity-led customer intelligence, GenAI cross-sell and a customer data platform spanning 15+ insurance products across Turkey and Lebanon.

  • Azure ML
  • Databricks
  • GenAI
  • CDP

~30% lead conversion

  • 20+ team
  • 3 regions
Illustration of rows of retail stores

Retail · Global

Pricing Analytics Modernisation

Delivered key workstreams for 14,000+ stores across Europe and North America: a 100TB+ Azure migration, scalable feature pipelines and downstream BI.

  • Azure
  • Databricks
  • Synapse
  • ADLS

14,000+ stores

  • 100TB+ Azure migration
Illustration of a magnifying glass over a signal chart

Medtech · APAC

Compliance ML Capability

Delivered ML compliance analytics across 12 APAC countries, lifting precision from ~70% to ~95%.

  • Python
  • Anomaly detection

~70% → ~95% precision

  • 12 countries
Illustration of data blocks moving from storage into a cloud

Financial services

500TB+ AWS Migration

Led delivery of a 500TB+ Redshift migration with ~$280K in annual savings.

  • Redshift
  • Glue
  • Athena

~$280K annual savings

  • 500TB+
Illustration of a network of connected entities under a spotlight

Insurance · Earlier

AML Detection & Reporting

Built an AML model that surfaces high-risk entities, and industrialised Power BI reporting for 200+ users.

  • Databricks
  • Power BI
  • Python

~80% cycle-time reduction

  • 200+ users
Illustration of friendly robots collaborating around a glowing sphere

Personal lab · 2025

AI Experiments & Product Thinking

Agentic workflows for intelligence gathering, profile matching and outreach automation; multi-agent orchestration and AI-assisted decisioning.

  • n8n
  • OpenRouter
  • Apify
  • MCP

Multi-agent orchestration

How I lead

Four stages take an idea from POC to P&L. Six moves are how I run the delivery along the way.

My rule

I don't do demos. I ship outcomes, with governance that holds up in the audit room.

  1. POC

    Frame it

    Ambiguous needs become use cases with success metrics, a roadmap and scope.

  2. Pilot

    Ready it

    Data readiness and solution direction, with unrealistic expectations challenged early.

  3. Production

    Govern it

    Governed, audited delivery: Responsible AI, model governance, human-in-the-loop.

  4. P&L

    Prove it

    Clear ownership and measurable outcomes: ROI, conversion, cost savings, adoption.

The six moves

  1. 01

    Business problem framing

    Turn ambiguous needs into AI/ML use cases, success metrics, a roadmap and a clear scope.

  2. 02

    Data readiness

    Profiling, data-quality validation, source alignment, entity resolution and clear business rules, before anything is built.

  3. 03

    Solution direction

    Shape AI/ML, GenAI, RAG, agentic and cloud-data architecture, and challenge unrealistic expectations early.

  4. 04

    Delivery governance

    Sprint governance, RAID logs, steering reviews, executive reporting and vendor coordination.

  5. 05

    Risk & controls

    Responsible AI, model governance, audit documentation, human-in-the-loop and production safeguards.

  6. 06

    Adoption & value

    Track ROI, conversion, retention, cost savings, productivity and real business adoption.

Lab work

Personal lab · Fictional data

Small tools and working demos I build in my own lab to test what it takes to run AI safely: guardrails, evals, governance and data quality. Everything runs on fictional data, and none of it is client work.

11 short films · 6 open-source tools · MIT licensed

Showreel poster: four tools for putting AI to work, safely
Showreel0:32 · Silent

Live demos7

  • Live demo

    Claims agent safety console

    An AI claims agent's actions, checked live against controls.

  • Live demo

    Prompt injection containment

    Replays an inbox of injected emails against an agent's controls.

  • Live demo

    Policy Q&A with citations and evals

    Insurance policy Q&A with cited answers, refusals when the documents say nothing, and evals.

  • Live demo

    PII masking lab

    Masks personal data in claim notes and measures what still leaks.

  • Live demo

    Pipeline data quality gate

    A data quality gate with schema, volume and drift checks, because most data incidents aren't model failures.

  • Live demo

    Is this message a scam?

    Checks a suspicious SMS or WhatsApp message for scam signs.

  • Live demo

    Policy fine-print scanner

    Finds waiting periods and limits in a health policy PDF.

  • Hugging FaceTry all seven demos liveOpen the demo hub

More demos to try:Claims triage deskWhat changedVoice note to tasks

Open-source tools6

  • Open source

    agentbreak

    Tests AI agents against attacks on fake tools and fails CI on unauthorised actions.

  • Open source

    ai-policy-lint

    Checks agent configs, MCP configs and system prompts against your company's AI policy.

  • Open source

    ai-register

    Scans a codebase for AI usage and keeps a reviewable register with draft EU AI Act tiers.

  • Open source

    prompt-ci

    Prompts as code: owners, versions, behaviour tests and a CI gate on prompt changes.

  • Open source

    mcp-gatekeeper

    A local MCP proxy that adds per-tool policies, human approvals and a tamper-evident audit log.

  • Open source

    local-redact

    Redacts sensitive data before you paste it into an LLM. Runs in your browser.

Check the fit

Project or role · Fit check

Paste a project brief or a role description. It's matched against my real experience: domains, regions, stack and how I run delivery, and it's honest about the gaps too.

Runs entirely in your browser. Nothing you paste is stored or sent anywhere unless you ask for the optional AI read.

–match

Paste a brief, or try the sample, to see a match score, strong matches and what's worth discussing.

Writing

Recent notes from my blog on taking AI into production.

All writing
Open to conversations

Stuck between POC and P&L?

Let's compare notes. I'm open to conversations about taking AI to production, transformation programmes and advisory work. I reply within 24 hours, straight from my inbox.

contact@mukherjeesuman.com
  • AI transformation programmes, from POC to governed production
  • GenAI, RAG and agentic AI delivery
  • AI governance, Responsible AI and CoE setup
  • Cloud and data modernisation for AI workloads

Active across India, the Gulf, Europe, APAC and global capability centres.

Send a message

A personal lab build on fictional data. Not client work.