I build products at the intersection of AI, software and real-world work.
I'm Ankit Kumar. Over the last decade and more, I've built products across healthcare,
enterprise software and semiconductors, from shipping SaaS products from zero to one to
building AI systems that operate on real business workflows. I started my career verifying
silicon at Broadcom and eventually moved into product, where I've spent the last few years
figuring out what it actually takes to make AI useful in production.
Today, I'm exploring what happens when we rethink software, and the way we work, around AI
from the ground up.
The model is rarely the hard part. The gap between a pilot that impressed the room and a
system the business actually runs on is where budget, patience and credibility get spent.
That gap is where most of my work has happened.
Selected work
What I've built
A few chapters from the journey, and the problems they taught me to care about.
Distyl AI · Applied AI · Enterprise · Healthcare
Contract intelligence for healthcare payers and providers
Led strategy, product definition and delivery of an LLM platform that reads payer and
provider agreements. The accuracy mattered less than the auditability: a human validation
layer let subject matter experts review, correct and sign off on model output, which is
what actually got it trusted and adopted. Also ran Fortune 50 engagements turning
ambiguous workflows into phased roadmaps with defined success metrics.
Also built prior authorization workflow agents. The focus there was the agentic workflow
itself: improving the core iteration loop so each pass produced real gains in accuracy and
precision, and so the agents stayed maintainable as the number of them grew.
97%+ extraction accuracy · 100,000+ documents
What I learned The hard part of enterprise AI isn't getting a model
to work once. It's building the context, evaluation, workflow and product around it so it
works reliably at scale.
Each dot is one document in a hundred. Three miss.
Datavant (Apixio) · Product · Healthcare AI · SaaS
AI as a service, and the platform underneath it
Owned the AIaaS Insights product from idea through launch, alongside the core SaaS
platform it sat on: SSO, user management, reporting. Most of the work was unglamorous
and load bearing, getting the platform fast and reliable enough that customers stopped
noticing it was there.
5x faster · 15% model gain · 20% CSAT lift
What I learned Healthcare has no shortage of data or AI potential.
The real opportunity is removing the friction between information, decisions and action.
Response latency, before and after.
Sattva · Product · SaaS · 0 → 1
Grant management for funders, and an app for health workers
Took Shift, a grant and program management platform, to market for CSR funders and NGOs,
owning go to market end to end. Then launched Maatr, an Android app for maternal
healthcare workers, built on field research across multiple locations rather than
assumptions made in an office.
25% user growth · INR 1.5M revenue in 8 months
What I learned Great products start with understanding the problem
deeply enough that the solution becomes obvious.
Revenue over the first eight months.
Broadcom · Semiconductor · Design Verification
Where the career started, verifying silicon
Started my career in semiconductor engineering, working on IC design verification across
five SoC projects and two business units.
5 SoC projects · zero escapes · 150+ verification algorithms
What I learned It gave me an appreciation for systems, precision
and the cost of getting things wrong, lessons that have surprisingly followed me into AI.
Five chips shipped. Zero escapes.
Building
What I'm building now
I'm interested in the next generation of software: products where AI isn't a feature bolted
onto an existing workflow, but part of the architecture from day one.
The interesting question isn't "How do we add AI to this?" It's "If AI could do the work,
what should this product become?"The question I keep returning to
I'm currently exploring and building around AI-native software, agentic workflows, evaluation
and the future of how people work with machines. Some of these experiments become products.
Some become essays. Some fail quickly. All of them are ways of figuring out what is actually
possible.
Twelve years spent moving up one layer at a time, and each layer left something the next one
needed. It is why a conversation about eval design and a conversation about margin can happen
in the same meeting.
Background
Strategy, down to the silicon
Read downward. Every layer rests on the one beneath it, and none of them were skipped.
Now
Founder & advisor
Building agentic AI for semiconductor engineering, advising on product strategy and applied AI, and convening the hardware-AI community.
Enterprise AI
Distyl AI · Applied AI
Worked alongside enterprise clients to design and ship applied AI into existing operations, systems and constraints.
Healthcare AI
Datavant (Apixio) · Product
AI-powered risk adjustment for healthcare payers and providers, in a domain where a wrong model output is a regulatory problem, not a bug.
0 → 1 SaaS
Sattva · Shift
Lead PM for a grant and program management platform, ideation through launch, owning business, sales, design and engineering. Processes tens of thousands of CSR assets and reaches 100,000+ beneficiaries monthly.
Consumer scale
OYO · Pricing and ML products
Led pricing and machine-learning products inside a high-growth consumer business, where model decisions moved revenue daily.
Silicon
Broadcom · Design Verification
Four years, five SoC projects shipped end to end across US and India teams with zero escapes, plus 150+ verification algorithms still used there as reference IP.
MBA, IIM Bangalore, with an exchange term at Texas A&M's Mays Business School.
B.Tech in Electronics & Communication Engineering, NIT Tiruchirappalli.
Writing and Contact
Writing, and conversations
Thinking out loud
I write about building with AI, product thinking, and the messy gap between what AI can
do in a demo and what it can reliably do in the real world.
I'm particularly interested in AI-native products, agentic systems, human judgment,
evaluation, and how people learn to work differently with intelligent software.
I share the things I'm learning, including the things that didn't work.
I enjoy conversations with founders, product leaders and builders who are wrestling with
something genuinely hard: a workflow that is painfully manual, a product that needs to
become AI-native, or a problem where it's not yet obvious what AI should actually do.
Bring me the messy version. I'll bring product thinking, AI experience and a willingness
to think from first principles.
