Selected work
How I think, drawn out.
Client names and confidential details are left out. I'm glad to go deeper in conversation.
Turning a state's diploma policy into a rulebook a machine can read
Can a student and their family see, today, how close they are to graduating?
A state redesigned its high school diploma: 42 credits across 7 subjects, readiness seals, and two honors tiers. The rules lived in policy documents. Families and counselors needed answers from real transcript data, and vendors needed the same answers through an API.
Key decision
The state owns the rules, so the rules live in data, not code. A policy change becomes a rulebook edit, and the engine never needs a release for it. The tool is framed as a planning aid, never an official determination.
Outcome
First client-visible deployment about a week after the repository was created. My own delivery analysis estimated 5 to 7 times less effort than a traditional build of the same scope. Production launch is targeted for early 2027.
How I did it
- Turned the policy documents into a published, machine-readable rulebook: 1 diploma, 6 seals, and 98 conditions across 18 rule types, plus the full course catalog and 45 career and technical pathways.
- Built a pipeline that regenerates the rulebook from the source documents, with a validator that rejects hardcoded conditions.
- Shipped a stateless evaluation engine and API driven entirely by the rulebook, backed by 43 automated scenario checks and a published API contract.
- Took the family-facing experience through four design iterations to sign-off by the client's CIO.
A playbook for turning client projects into products
Build with clients, not ahead of them.
A services firm earns its revenue one custom project at a time, and most of that work is never used again. I wanted every engagement to leave something reusable behind, without asking the business to fund speculative products.
Key decision
Clients fund the build; the company funds only the branding, documentation, and hardening. Nothing gets branded until it has worked in at least two client implementations.
Outcome
A family of reusable products grew out of client work, including a configurable workflow engine, a data import and validation tool, and MASTERed, a governed master data platform I conceived and took from whiteboard to production over three years. Identity and finance interoperability frameworks are in the pipeline.
How I did it
- Defined a three-stage maturity path: a custom project, a reusable platform abstracted from it, and a branded, documented, supported product.
- Set a five-part screen for any candidate: repeatability across clients, pain intensity, domain fit, extendability, and ease of packaging.
- Built it into delivery. Product managers and analysts flag reusable patterns during requirements, teams generalize existing code into prototypes, and new clients see them as proven accelerators.
- Added governance: a recurring productization review, a shared backlog, and one or two components packaged per quarter.
Rebuilding a product team's delivery process around AI
Requirements were the bottleneck: slow to write, slow to clarify, and expensive to fix after release.
I lead a cross-functional team of five product and business analysts supporting state data platforms, and a CTO-sponsored program to bring AI into every stage of the software lifecycle.
Key decision
Sensitive client data stays local. Privacy-sensitive work runs on a local retrieval setup, so the team gets AI help without sending client data out.
Outcome
Faster requirements turnaround and fewer changes after release. On the flagship diploma platform, the same approach put a client-visible build in front of the state about a week after the repository was created.
How I did it
- Planned a catalogue of about 52 AI tools across analysis, product, development, testing, DevOps, architecture, and reporting.
- Piloted an AI backlog manager that turns meeting transcripts into proposed backlog changes, applied only after a person approves them. Deployed on Azure with managed identity.
- Rolled out adoption by use case, starting with three concrete uses per person rather than a general mandate.
- Measured AI-assisted delivery against a traditional baseline, with evidence rather than anecdotes.
Personalized card acquisition at American Express
A card application should feel like it was built for the person filling it in.
As Technical Project Manager for American Express, through Cognizant, I worked on card acquisition, where generic offers and landing pages left value on the table.
Key decision
Two audiences, two signals. Existing members saw offers tied to their rewards; new visitors were matched to a card from their browsing behavior.
Outcome
A launched small business card, and an acquisition experience personalized from the first page through the application.
How I did it
- Led the launch of a small business credit card from requirements through release.
- Delivered an application journey that presented member-specific Membership Rewards points offers.
- Built landing pages that matched each visitor to the most relevant card using cookie data.
Career
18 years, drawn to scale.
- 2022 to nowProduct Manager · EdWise Group
- 2017 to 2021Product Manager · Arizona Department of Education
- 2014 to 2017Technical Project Manager · Cognizant, for American Express
- 2011 to 2013Internet Banking Product Owner · ICICI Bank
Executive MBA · University of ArizonaProduct Manager Certificate · Product SchoolB.E., Information Technology · University of Mumbai
Contact
hello@aadityahirurkar.com
Glad to connect with anyone working on data, interoperability, and AI in regulated industries.