Bank-grade reliability. Startup speed. Shipping platforms that move money and data, and measuring every step.
Seattle, WA · Sr. Product Manager · Fintech & Platforms

Engineer →
Product Leader.

Building payment APIs, developer platforms & agentic AI that move money and data at scale.

Building from the rails up. Every integration, checkout and claim should take less time than the last. 7+ years across fintech, B2B SaaS and AI at ICICI Bank, Microsoft, EY and DoctorAlliance. From India's first Banking-as-a-Service platform (400+ fintech partners, 166B transactions a year) to B2B checkout at Microsoft and agentic AI in healthcare today. It started as an electrical engineer on industrial IoT, where uptime was the product. Every product since has lived in a regulated, always-on industry, so reliability comes first.

Dhrub Chand Thakur Fintech · Commerce · AI Platforms
2015 MachinesIoT telemetry, predictive maintenance TATA Voltas
2020 DataSelf-serve analytics, pipelines EY
2021 APIs & money railsBanking-as-a-Service platform ICICI Bank
2024 CommerceB2B checkout, billing, provisioning Microsoft
2025 AgentsAgentic document intelligence DoctorAlliance
Each layer sits on the last Open to roles
7+
Years in product · payments, platforms & AI
166B
Transactions a year · ICICI BaaS platform
400+
Fintech partners across 29 markets
38→26%
B2B cart abandonment · Microsoft
14d→48h
Partner integration time · ICICI
7d→<4h
Claims processing · agentic AI
The through-line

Three habits I carry into every product.

01Chillers → bank rails → claims

Reliability is the product.

I started where equipment uptime was the product: a chiller that stops can stop a customer's plant. Every industry since has been regulated and always-on, from bank rails at 99.997% accuracy to healthcare claims. So I plan the fallback, the alert and the audit trail before the feature. Available and trusted comes first; fast comes second.

02Telemetry → API data → agents

Instrument before you decide.

A fleet of chillers told me which configurations failed in which climates. Years later, a strange pattern in API data (every payment worth exactly $1) told me partners needed a validation product. The data usually knows first.

03183 SKUs → 35 · one API

Simplify the platform, not the promise.

Collapsing 183 controller variants into 35 modular ones is the same move as giving partners one payment API that quietly reroutes across four rails. Complexity moves inside; the customer sees less of it.

Selected Work

Six products, one layer at a time.

Newest first. Each study walks the context, the problem, the approach and the outcome, then names what I carried forward into the next chapter. Open any card.

Context
A healthcare SaaS platform handling 50K+ orders a month, where medical claims took more than seven days to process.
Problem
Claims throughput was capped by how fast information could be pulled out of medical documents accurately. More people would not fix it, and a model that was fast but wrong would make it worse, because errors in claims cost money and trust.
Approach
Owned roadmap and execution for an agentic AI document-intelligence platform.
  • Wrote the PRDs and defined success metrics around both speed and extraction accuracy, so neither could be traded away quietly
  • Set delivery milestones across Engineering, Data Science and Operations
  • Treated accuracy as the product: the agent had to be trusted before it could be fast
Outcome
Claims processing went from 7+ days to under 4 hours, with extraction accuracy up 60%.
Carried forward from TATA VoltasAn agent is a sensor with judgment. The same rules apply: measure precision, not just speed, and design for what happens when it is wrong.
Agentic AIDocument IntelligenceHealthcare SaaSPRDs & Metrics
Context
Business customers buying through the Microsoft Store and the M365 / Azure Admin Center. 38% abandoned the cart, and license provisioning took about seven days of largely manual work.
Problem
Two leaks in one funnel: buyers dropped before paying, and the ones who paid waited a week to use what they bought. Acquired companies also could not inherit their parent's contract pricing without manual intervention.
Approach
  • Redesigned checkout into a 3-step progressive flow and ran multivariate experiments across 50K+ users
  • Introduced smart-bundling recommendations inside the purchase path
  • Shipped a self-serve billing and payment-provisioning experience, including automatic parent-contract price inheritance for acquired entities
  • Pushed back when the team felt directory validation alone was enough for entity mapping, and championed a maker-checker, bidirectional approval step on existing access controls. It closed a real fraud path at little added build cost
Outcome
Abandonment fell from 38% to 26%, time-to-purchase halved, and 85% of manual provisioning work disappeared.
Carried forward from ICICIMoving money taught me that the approval flow is part of the product. The maker-checker call came straight from bank-grade habits.
CheckoutExperimentationBillingB2B SaaSTrust & Safety
Context
A $100B+ bank opening its payment rails to fintechs through APIs, SDKs and developer tooling. I led the platform with a team of 3 APMs, 2 engineers and 2 analysts.
Problem
Partners took 14 days to integrate, regulated SMB-facing partners in insurance and lending needed near-zero failure, and the API portfolio had not grown into the use cases partners actually had.
Approach
  • Launched a developer portal with self-serve onboarding, sandbox testing, SDKs and documentation
  • Designed Composite PAY, which reroutes failed payments across NEFT, IMPS, UPI and RTGS without the end user or partner noticing
  • Expanded the portfolio into trade finance, supply chain, BNPL, investments and account opening; shipped smart routing, beneficiary validation and embedded collections
  • When a large broking client demanded a real-time callback for every status change, proposed a 10-minute batch update on pending transactions instead. They accepted, and it became the base of our bulk-payment product
Outcome
$20M+ platform revenue, 166B transactions a year at 99.997% accuracy, and 400+ fintech partners across 29 markets. Composite PAY lifted success in high-failure scenarios by 15% and held through two tax-season surges with zero downtime incidents.
Carried forward from TATA VoltasComposite PAY is the chiller retrofit in software: assume a component will fail, and make the fallback invisible to the person relying on it.
0-to-1 PlatformBaaSDeveloper ExperiencePayment RoutingRegulated
Context
Two small signals. In the API analytics, some clients' total transaction value equaled their transaction count: every payment was exactly $1. Separately, sales reps kept pulling the product team into deals for basic solutioning and compliance questions.
Problem
Partners were sending $1 test payments to check a beneficiary before moving real money, which loaded our rails and cost them time. Sales, meanwhile, moved only as fast as product could answer them.
Approach
  • Checked the aggregation logic and raw data, then interviewed clients to confirm the hypothesis before building anything
  • Shipped a real-time, non-monetary Beneficiary Validation API, later chained into a register → validate → pay flow
  • Built an AI sales-enablement platform: a client-aware product recommendation engine plus a RAG chatbot for compliance and integration questions, piloted by zone with feedback loops
Outcome
Validation reached 70% client adoption in six months and became a new revenue line. The sales platform improved deal win rates 30% and cut the sales cycle from 12 days to 3.
What I would do differentlyTrack the validation flow's effect on payment volume from day one, so any cannibalization shows up in the data and not in a quarterly review.
0-to-1RAGRecommendation EngineAnomaly → Product
Context
Data and digital transformation for Fortune 50 clients. The lead engagement was a self-serve analytics platform for a global CPG firm with 5,000+ internal stakeholders.
Problem
Decisions waited about 24 hours for someone to pull and shape the data. The people making the call could not get the answer themselves.
Approach
Led discovery, roadmap and delivery, and owned the whole data lifecycle underneath: ingestion, cleaning, load, business rules, validation, and the Power BI layer on top.
  • During a COVID wave, two analysts on a supply-chain data-lake deadline fell sick. I taught myself Databricks debugging and fixed the critical lines in a 2,000 to 3,000 line Python/SQL codebase in two days. We shipped on time and the client expanded scope to more geographies
Outcome
82% weekly active usage across 5,000+ stakeholders, and decision time cut from 24 hours to 60 minutes.
Carried forward to ICICIAdoption is the metric. A dashboard nobody opens is a sensor nobody reads.
Data ProductSelf-servePower BIDatabricks
Context
Tata's $6B industrial equipment division maintained chillers at 1,000+ customer locations across India, on 25-year equipment lifespans with a new generation every three to four years. Many generations ran side by side in the field.
Problem
Maintenance ran on a calendar. Diagnosing a fault often took multiple site visits, and nobody collected structured failure data, so nothing fed back into product design.
Approach
Led the pivot from preventive to predictive maintenance, working with TCS as technology partner on an IoT gateway and big-data stack.
  • Ran partner selection, field configuration and a 40-unit pilot, then scaled across a mixed-generation install base
  • Shipped alerting in two stages: threshold rules routed to the mapped service engineer first, ML alerts trained on fleet patterns second
  • After 12 months of data, correlated performance with ambient temperature, geography and duty cycle to see which configurations suited which conditions. That fed next-generation design and gave sales a region-specific guide
  • Found 183 controller SKUs behind slow service, and redesigned them into 35 modular, cross-compatible SKUs
Outcome
5,000+ connected units at 70% activation and 91% monthly active use, 40% less downtime, 86% alert accuracy. The SKU redesign cut inventory holding cost about 60% and service turnaround 30%.
The seed of everything afterThis is where I learned the loop I still run: instrument, find the pattern, change the product, and make sure the people in the field actually act on it.
IoTPredictive MaintenanceML AlertingHardware PMPlatform Simplification
Career Arc

From plant rooms to platforms, in six chapters.

I did not switch careers; I kept moving up the stack. Each chapter added a layer: machines, then data, then money rails, then commerce, then agents. Two degrees in the middle gave me the finance and analytics vocabulary to make the jump from hardware to software products.

2015 – 2018 · Ch. 01
Hardware & IoT PM
TATA Voltas

Electrical engineer on commissioning sites, then product owner for connected chillers, field-failure capture and a spares ordering portal.

Added: instrumentation, field empathy
2018 – 2020 · Ch. 02
PGDM: Finance, Operations & Analytics
IIM · Dean Scholar (Top 5%)

Went back to school for three majors (finance, operations and analytics) to learn the language of money and data before building products about both.

Added: finance, analytics
2020 – 2021 · Ch. 03
Senior Consultant, Digital Products
Ernst & Young

Self-serve analytics for a $128B CPG firm: 82% weekly active use across 5,000+ stakeholders.

Added: data products, enterprise buyers
2021 – 2023 · Ch. 04
Senior PM, API Banking
ICICI Bank

India's first BaaS payment platform: 400+ fintech partners, 166B transactions a year, $20M+ revenue. First production AI with a RAG assistant.

Added: platforms, APIs, payments
2023 – 2025 · Ch. 05
MBA · Microsoft
UW Foster, Top 5% · Sr. PM (MBA Intern)

B2B checkout and self-serve billing across M365 and Azure: cart abandonment from 38% to 26%, license provisioning from 7 days to 4 hours.

Added: SaaS commerce, experimentation, AI projects
2025 → Now · Ch. 06
Senior Product Manager
DoctorAlliance

Agentic AI document intelligence for healthcare: claims from 7+ days to under 4 hours, 60% better extraction accuracy.

Added: agentic AI in production

Hardware taught me that you cannot patch a machine after it ships, so you design for failure and instrument everything. Software let me run that loop a thousand times faster. AI is the first time the product itself can read the sensor.

Projects · github.com/chanddhrub

AI projects, built end to end.

Work outside my day job where I own the whole loop: research, PRD, pricing and build. Each one carries its honest status, from shipped MVP to design stage.

AI projects 2 projects · LLM agents · Next.js

Why this shelf matters: both start from the question every AI buyer asks first: can I trust the answer? One raised retrieval accuracy before anything else; the other refuses to show a claim it cannot quote.

AI agent · Natural language to data · 2024 Shipped MVP

Natural-language data query agent

Built and launched the MVP of an AI agent that answers business questions in plain English against an enterprise's own data, working with engineering and customer teams. Then designed its tiered pricing and bundling by comparing 7+ AI model alternatives on cost and quality.

+20% retrieval accuracy+30% revenue potential7+ models compared

Pharma and sports clients got more accurate answers, and cheaper model variants opened global SMB segments.

AI · Legal docs · Next.js · Supabase Design stage

Redline: read the offer letter before you sign it

Upload an offer letter, non-compete or IP assignment and get the risky clauses ranked by how hard they are to escape, each with the exact sentence it came from and a drafted counter. A flag without a quotable source is treated as a bug and dropped.

PRD + ADRsResearch repoCitation-first AI

An AI product designed around verifiability, with its costs (lower recall, no OCR) written down on purpose.
Fintech & go-to-market projects 2 projects · React Native · Strategy

Why this shelf matters: one is a consumer fintech app I am building now; the other brings my hardware roots back in, taking a predictive-maintenance sensor from lab to market.

Fintech · React Native · Expo · Plaid In build

Fina: privacy-first transaction analytics

A cross-platform mobile app for spending, subscriptions and budgets, where all transaction analysis happens on the device. A Plaid-connected server fetches data; nothing is stored on it. The privacy-first alternative to Rocket Money-style apps.

iOS + AndroidPlaid linkOn-device analytics

Tests a real consumer bet: people will trade server-side features for keeping their data.
Deep tech · IoT sensor · MBA, Creative Destruction Lab Delivered

Go-to-market for a predictive-maintenance sensor

A deep-tech startup had a peel-and-stick sensor backed by 27 patents but no segment focus or pricing. I sized a $5.1B to $24B market, ranked 5 verticals by adoption readiness from 8 structured interviews, laid out two GTM paths (direct to plant vs. OEM licensing), and drafted hardware-plus-SaaS pricing with an ROI model.

3 new pilot talks1 OEM proof of concept5 verticals ranked

The hardware PM and the SaaS PM in one project: sensor economics priced as a subscription.

Toolkit

The skills stack, and where each one was earned.

Every skill is tagged with the place I used it. Read top to bottom and it tells the same story as the career arc: hardware at the base, AI at the top.

◆
AI & AgentsSince 2021
Agentic document intelligence
DoctorAlliance
RAG chatbot · recommendation engine
ICICI
ML alerting on sensor telemetry
TATA Voltas
Natural-language data query agent (MVP)
Project
AI pricing & model cost trade-offs
Project
Citation-first, verifiable AI design
Redline
★
Commerce & GrowthSaaS · B2B
Checkout & conversion optimization
Microsoft
A/B & multivariate experimentation
MicrosoftICICI
Billing & self-serve provisioning
Microsoft
Pricing, packaging & bundling
MicrosoftProject
Self-serve analytics adoption
EY
Market sizing & 0-to-1 GTM
Project · MBA
⌥
Platforms, APIs & PaymentsRegulated
Banking-as-a-Service & API portfolio
ICICI
Developer portal · SDKs · sandbox
ICICI
Payment routing & fallback (UPI, IMPS, NEFT, RTGS)
ICICI
Wallets, loyalty & bill-pay mandates
ICICI
Maker-checker & access controls
MicrosoftICICI
Partner onboarding at scale
ICICI
⚙
Hardware & IoTThe foundation
IoT telemetry & remote monitoring
TATA Voltas
Predictive maintenance (rules → ML)
TATA Voltas
Field-failure capture & root cause
TATA Voltas
Modular design · SKU rationalization
TATA Voltas
Technology partner management (TCS)
TATA Voltas
Salesforce CRM + SAP data joins
TATA Voltas
Tools, languages & methods
SQL Python R · R Shiny Databricks Power BI Azure ML Studio Jira · Agile / Scrum Salesforce · SAP PRD / BRD · RICE Causal inference · funnel & cohort analysis NPV · ROI · market sizing
Résumé

The one-pager, in full.

The same story as a traditional résumé, print-ready. Download it, open it in a new tab, or read it inline below.

Download PDF Open in new tab Reply by email
Experience
  • DoctorAlliance · Senior PM (2025 → now)
  • Microsoft · Senior PM, MBA Intern (2024)
  • ICICI Bank · Senior PM, API Banking (2021–2023)
  • Ernst & Young · Senior Consultant, Digital Products (2020–2021)
  • TATA Voltas · Associate PM, IoT & SaaS (2015–2018)
Signature work
  • Agentic AI claims platform, 7+ days to under 4 hours
  • India's first BaaS payment platform, $20M+ revenue
  • Microsoft checkout redesign, abandonment 38% to 26%
  • Developer portal, integration 14 days to 48 hours
  • IoT predictive maintenance, 40-unit pilot to 5,000+
Education & credentials
  • MBA, UW Foster · Dean Scholar, Top 5%, STEM (2025)
  • PGDM, IIM · Finance, Operations & Analytics · Dean Scholar, Top 5% (2020)
  • BE Electrical Engineering, NIT (2015)
  • LLM Foundations Certified PM · AI Product Management
  • API Product Manager · Lean Six Sigma Green Belt

Your browser can't show the PDF inline.

Open the résumé
Contact

If you are building where money, platforms and AI meet, and you want a PM who has shipped at every layer from sensor to agent, I would like to hear about it.

chanddhrub.thakur@gmail.com 425 · 535 · 8406 LinkedIn GitHub
Based in
Seattle, WA
Open to
Relocation within the US
Best for
Senior PM · Lead PM
Focus
Fintech · Commerce · AI platforms