0
Initializing
AI Product Manager · Pune, India

Builds AI products people actually need.

I grew up watching products win and fail in my family's retail shop — now I build them, deliberately, in AI. 4 shipped AI products and counting.

Scroll
AI Product Manager·4 AI Products Shipped·RAG Pipelines·AI Evals·Voice of Customer·Ships, Not Slides·
01

From a family shop to AI products

I grew up watching products win and fail in my family's retail shop, long before I had words like "product-market fit."

I'm moving into AI Product Management deliberately — not because it's trending, but because it's the most important product shift of our generation.

The last year: building XR and VR products at Firebird VR in Pune. Before that: founded UnmuteX, a mental wellness platform, during COVID. My path runs through XR, UX and game design — and it converges on AI product.

Gowtham Kumar
Gowtham Kumar · Hear my intro ↑
4
Shipped AI products
16wk
LLMs, RAG, prompt engineering, AI evals, product strategy
1yr+
Building XR/VR at Firebird VR
0→1
Founded UnmuteX during COVID
02

Selected work

/01

TruthLens Case study

AI deepfake detection for the people most likely to fall for one — non-tech-savvy parents in Tier 3/4 India — not the IT teams who already know how to spot one.

500M+ target users Free for parents · ₹99/mo guardians 96.5% accuracy target
Two users, one product
  • The Parent — protected, one-glance answers
  • The Guardian — pays, gets instant alerts
Honest risks

A missed scam is worse than a false alarm. Scammers will study it back. DPDP Act consent is genuinely hard.

/02

VoC Copilot Live

Paste raw customer feedback → AI clusters it into themes, attaches real user quotes, flags churn risks and recommends roadmap priorities. Tested on real Ludo King review data.

Every insight cites its source Built-in eval framework: good vs bad AI output
0
/03

PM Resume Tailor Live

AI tool that analyzes the gap between your resume and a job description — built for aspiring PMs in India.

/04

PrepTalk Live

A voice-first AI interview coach — rehearse out loud and get real feedback, not tips on a page.

03

Stack & concepts

Tools I ship with

ClaudeCursorLovableBoltStreamlitLangChainRAG pipelinesPrompt EngineeringFigmaChatGPTGoogle Gemini

Concepts I work in

LLMsRAGAI EvalsHallucination guardrailsAI product strategyVoice of Customer analysis
04

How I think

01

Build for the person most likely to fail

TruthLens isn't for IT teams who spot deepfakes — it's for the 65-year-old parent in Bhadrachalam who can't tell a fake call from a real one. If the most vulnerable user can use it, everyone can.

02

Every AI output must cite its source

VoC Copilot doesn't just cluster feedback — it shows you exactly which user said what. No black boxes, no "the AI thinks." If it can't cite it, it doesn't ship.

03

Define "good" and "bad" output before building

Before writing a single prompt, I write the eval framework: what does a correct answer look like? What does a dangerous one look like? A missed scam is worse than a false alarm — that shapes everything.

04

Ship, then refine — not the other way around

Four products in four months. None started perfect. PM Resume Tailor launched in a day, got real user feedback, and improved. The best PRD is a live product with actual usage data.

Right now: preparing for APM interviews, deepening AI/ML fundamentals — targeting offers by September 2026

Let's build
something real.

If you're hiring, building, or just want to talk product — I'd love to hear from you.

🥚✨
You found the easter egg!
Fun fact: The first product I ever "shipped" was a handwritten price list for my family's shop when I was 12. It had columns, categories, and a loyalty discount system. Some things don't change.