00 Portfolio · 2026

Building AI-native products,grounded in real business problems.

Engineer turned product builder. I work at the intersection of AI systems, product thinking, and enterprise workflows, on the seam where model capability meets the way teams actually get work done.

See selected work Start a conversation

01Selected Work

Three projects at the seam of AI capability and enterprise execution.

PRJ 01Agentic Systems2025, Ongoing

Multi-agent enterprise workflow system

An orchestration layer for AI agents that can be trusted with real work, not just demos.

Request walkthrough

Problem

Single-agent LLM patterns break down on enterprise workflows that span multiple tools, long contexts, and branching decisions. Most agent frameworks ship as toys; they do not survive contact with a real ticketing system or a finance reviewer.

What I built

An orchestration architecture that routes work between specialized agents with explicit tool contracts, persistent memory, and a coordination layer that handles retries, hand-offs, and human checkpoints. Designed for reliability first: observable, auditable, and degradable.

Stack

TypeScriptLangGraphLangChainVector DBsLLM APIsMCP

Impact

Reliable tool routing on long-running workflows that previously needed constant human babysitting.

PRJ 020 to 1 Product2025, Closed beta

TheBookShelves

A community network for readers, hosted inside independent cafes.

See the prototype

Problem

Two compounding problems on either side of one room. Independent cafes face thin margins and quiet off-peak hours; readers have nowhere community-rooted to land between home and work. Discovery is broken on both sides.

What I built

An AI-native, two-sided platform. Partner cafes host curated bookshelves and reading rituals: silent reads, themed nights, slow mornings. Readers find cafes, browse books, reserve titles, and join events that match the way they want to be in public.

Stack

Claude CodeLovableSupabaseNext.jsMiro

Impact

Working prototype shipped; closed beta planned with partner cafes in Bengaluru.

PRJ 03Strategy & Research2025

AI Monetization Framework

Why most enterprise AI pricing quietly loses money, and what to do about it.

Request the paper

Problem

Enterprise AI pricing is being copy-pasted from SaaS playbooks that do not account for non-linear cost-to-serve, fragile adoption, or who actually captures the value when an outcome is generated.

What I built

A framework that maps pricing across SaaS copilots, usage-based, outcome-based, and embedded models. Benchmarked 15+ products and built an analytical lens for evaluating any AI pricing decision against cost-to-serve, value capture, and adoption friction.

Stack

Market researchUnit economicsExcel / SQLComparative analysis

Impact

A repeatable lens used in 3+ pricing conversations across academic and enterprise contexts.

02About

Three years inside enterprise systems. That is where I learned to respect the work.

Most AI products fail at the seam between model capability and the way a real team actually gets work done. I build for that seam.

I started as an engineer at Capgemini, running reliability and observability for retail, pharmacy, and hospitality platforms used by Fortune 500 clients. I learned what production really means: that a 30% latency improvement matters because someone, somewhere, is trying to close a register.

At MathCo I moved closer to the product surface. I built a GenAI copilot for report generation that shipped across 5+ enterprise customers, and I owned the collaboration layer of an analytics platform: comments, tasking, attachments, the unglamorous primitives that decide whether a team actually adopts your software.

The MBA at IIM Udaipur in Digital Enterprise Management gave me the vocabulary for the business questions I was already asking on the floor: pricing, adoption, unit economics, organizational change.

Now I build agentic systems and AI-native products with the same instinct. Start from the workflow, not the model. Let the work speak.

2025 — Now

Building / TheBookShelves & agentic systems

Closed beta with partner cafes in Bengaluru. Multi-agent orchestration on the side.

2025 — 2026

MBA, Digital Enterprise Management / IIM Udaipur

Focused work on AI monetization, enterprise adoption, and product strategy.

2024 — 2025

SWE (Product) / MathCo

Built GenAI copilot for report generation, shipped to 5+ Fortune 500 clients. Owned collaboration layer.

2022 — 2024

Software Engineer / Capgemini

Reliability & observability for retail and pharmacy platforms. Failure detection: hours to under 5 minutes. 1,000+ RCAs.

2020

B.Tech, Mechanical Engineering / Alliance University

03Principles

How I think about building products with AI inside them.

Start with the workflow, not the model.

The question is not what the model can do. It is which step in this human's day is brittle, and would a model make it less brittle.

Reliability is the feature.

A clever agent that fails 1 in 10 times is a worse product than a boring one that fails 1 in 1,000. Enterprise teams trade flair for trust every time.

Adoption is an interface problem.

The collaboration layer (comments, tasking, attachments, review) is where AI features earn a place in the workflow or quietly die.

Price for cost-to-serve.

SaaS pricing on a product whose marginal cost moves with inference tokens is a quiet way to lose money. Unit economics belong in the pricing conversation.

Observability before cleverness.

If you cannot see every tool call, prompt, and fallback, you cannot improve the system. Logs and traces are how the product gets better.

Ship the spec, then the system.

A crisp written spec (what the agent does, refuses, escalates) is the actual product. Most failures are spec failures wearing engineering costumes.

04Writing

Notes from the seam. Short essays on AI products, pricing, and adoption.

May 2026Why your AI copilot keeps getting unused, and the collaboration layer underneath it Apr 2026The four ways to price an AI product, and why three of them quietly lose money Mar 2026Agents do not fail at reasoning. They fail at hand-offs.Feb 2026Building TheBookShelves. Designing for ritual, not for scale.

05 Contact

If you are building something hard at the AI / product seam, let's talk.

Open to PM, AI PM, Product, and BA roles where business, AI, and engineering judgement actually meet. Also: collaborators on TheBookShelves, and anyone thinking carefully about agentic systems in production.