How does an AI demo become a product people trust?
I am exploring what changes when an AI system has to work repeatedly, safely and usefully outside controlled conditions.
Product leader · Former founder · Builder
I build AI products. And rethink how we build them.
I have spent the last 13 years building products, including a company of my own. Today, I build AI products and study what it takes to turn an impressive demo into something people can genuinely use and trust.
Building for a probabilistic world
Noida, India · 2026
Explore
Now / 2026
BuildingAI products, agents and shared platforms
LearningAI engineering, evals, voice and Python
WritingHow probabilistic products get built
ExploringTechnology, work and human decisions
01 / Thinking
The questions I keep returning to.
I keep seeing the same pattern. An AI demo works beautifully in a controlled setting, but the real world is messy. Context is incomplete, answers vary and trust is easy to lose. That has made me question many of the practices we use to build and launch products.
How can AI find what truly matters inside an organization?
Enterprise knowledge is scattered across documents, conversations, applications and people. Finding an answer is only part of the problem. Knowing whether to trust it is harder.
What evidence is enough to launch an AI product?
Traditional QA gives us pass or fail. AI asks us to reason about quality, risk, confidence, cost and behavior that may change from one interaction to the next.
Which problems are actually worth building for?
Speed matters only after we choose the right problem. I am interested in the decisions, blind spots and assumptions that come before the roadmap.
02 / Work
Things I am building and exploring.
These are not finished answers. They are the ideas, lessons and experiments currently shaping how I build.
SPIRAL
A practical method for answering a question I kept encountering: if an AI demo can be convincing and still be wrong, what should teams do before calling it ready?
Making organizational knowledge useful
What several years of building search taught me about relevance, permissions and why finding the right answer is ultimately a problem of trust.
Learning AI by building
I do not want to understand AI only through roadmaps and diagrams, so I am learning the technical layer by building, breaking and documenting things myself.
03 / Writing
Ideas that deserved more than a post.
Occasional notes on AI, products, building and what I am learning along the way.
Why AI products break traditional SDLC
Requirements, logic and expected outputs worked when software was deterministic. AI changes the contract.
Read the essay ↗When everyone’s ready to build, who is asking if we should?
AI has made execution faster than ever. But speed helps only after we decide what deserves to be built.
Read the essay ↗From search to knowledge discovery
A practical look at the trust, context and retrieval foundations behind enterprise answer experiences.
Read the essay ↗Stay with the longer ideas.
One thoughtful essay at a time. No content treadmill.
05 / About
A builder, in different forms.
I have spent more than a decade building products. Some years were spent inside companies, and some were spent trying to build a company of my own.
The titles and industries changed, but I kept gravitating towards the same kind of work: unclear problems, zero-to-one journeys and ideas that needed someone to bring structure without removing the possibility.
Being a founder taught me to care about survival and distribution, not only product quality. Working inside larger organizations taught me how difficult it is to turn a good idea into something that teams can align around, build and scale.
I am now beginning another chapter, going deeper into AI products and the systems behind them. This site is where I intend to document that journey honestly, including what works, what breaks and what changes my mind.
06 / Speaking
Talks and conversations.
Speaking helps me pressure-test an idea in public. I occasionally talk about AI products, enterprise search and how the craft of building software is changing.
Beyond Deterministic Product Management
Why AI products break traditional SDLC, and what product teams need instead.
Ask about the talk ↗Make Search Work Better for Employees
How better discovery connects employees with the knowledge they need to move work forward.
Invite me to speak ↗Continue the conversation
Let’s stay connected.
If these are questions you care about too, LinkedIn is where I share the shorter thoughts and Substack is where I stay with an idea for longer.
Shorter thoughts. Open conversations.
I use LinkedIn as a working notebook: a place to share ideas before they become essays, learn from disagreement and document what building is teaching me.
Why do strong AI demos still fail in the real world?
Because a happy path proves possibility. It does not yet prove reliability, trust or readiness.
Explore my LinkedIn ↗Finding an answer is becoming a trust problem.
Retrieval matters, but permissions, context, freshness and the reason behind an answer matter just as much.
Explore my LinkedIn ↗Going deeper than the product layer.
Notes from learning Python, agents, evals, voice AI and the systems underneath the product decisions.
Follow the journey ↗