rushabh-parikh/portfolio.board

You

Hello

Rushabh Parikh
the headmaster

AI Product Manager · Jumbo Homes · Bengaluru

Rushabh
Parikh.

I build AI products and own the numbers they move.

receipts tap one for the proof

I lead AI products at Jumbo Homes, where our WhatsApp agents sell real homes end to end: they list the home, market it, answer buyers and book the visits. Before that: deep tech at Chakr and core engineering at OLA Electric.

Roadmap · 2018 → now

Four years, three industries, zero boring ones

Click a bar

Study

Core engineering

Deep tech

Product & AI

Jul 2025 – now · Chief of Staff → Product

Jumbo Homes

Joined the founding team as a generalist and became the product owner. Shipped two customer-facing AI agents and a crew of internal ones. ARR grew about 5× in a year.

OLA Electric · 2022–24

Assistant Manager, Vehicle Engineering

Taught a model to crash cars, so the supercomputer didn't have to

Every new car must prove it won't badly injure a pedestrian in a crash. Each check was a simulation that ran for hours on a supercomputer. I built an ML model that predicts the injury score up front, so engineers only run the full simulation on designs worth testing.

500 hrscompute saved per design iteration

₹1.7 Crsaved in operating costs

Hover a point. Illustrative data fitted to the model's R² of 0.83.

Lab notebook · Chakr Innovation · 2024–25

Technical Project Manager, Moonshot R&D

Tried to power EVs with aluminium and thin air

An aluminium-air battery makes electricity from aluminium metal and oxygen in the air. It's cheap and packs a lot of energy, but the metal also corrodes, wasting itself as hydrogen gas. I led the six-person R&D team that cut the waste and the cost.

−1.35 V

Drag right: a more negative anode potential means less aluminium lost as H₂ and a more efficient cell. Illustration, not lab data.

  • 83%of peak performance from scrap aluminium, the key to commercial viability
  • −40%raw-material cost. I built the supply chain on the ground in China
  • 5× fasterand 90% cheaper testing, using an ML model I built
  • Amazon, Swiggyinterested after a month-long vehicle pilot. I ran the go-to-market myself

On WhatsApp · Tara + Ananya

Live PRDs, prompts, evals and a fair bit of the code: me

Two agents, one home, zero apps

Tara works the sell side. Ananya works the buy side. Both live on WhatsApp, and between them they've sold real homes end to end.

Tarasell side · WhatsApp online

  1. Hi! I'm Tara. Tell me about the home you want to sell. No forms, promise.

Chat with Tara on WhatsApp ↗

Tara · sell side

  • 28× faster onboarding: 3.5 days → ~3 hrs from first message to live
  • −32% latency after restructuring tools and skills
  • Vibe-coded on a product that barely existed. PRD: mine.

Ananya · buy side

  • 2× conversion: enquiry → booked meeting, 4.5% → ~10%
  • Leaks fixed: rigid scheduling, irrelevant recommendations, weak visuals
  • Evals: an LLM judge asks "did it do its job?"

scripted replay · time compressed

Ananyabuy side · WhatsApp online

  1. Hi! Ask me anything about a home, or pick a time to see it. Any time. I don't do slots.

Chat with Ananya on WhatsApp ↗

Dashboard · Jumbo Homes, year one

One year. Receipts attached.

Grey is where it started. Colour is where it got to a year later.

  • ARR

    ₹1.15 Cr₹6 Cr5×

  • Deals closedper month

    4215×

  • Live inventoryin one quarter

    ~100800+8×

  • Marketskept only the strong ones

    3113.7×

4× deals per head vs. a traditional competitor, from internal tools I built.CM3-positive sales comp that still pays above market.

The crew · internal agents

Meet the crew

tap a card to flip it

Agents I built so the team spends its time on judgement, not grunt work. Names chosen with great care and zero focus groups.

Built by me with Claude · GPT · Gemini · n8n · Make · Python

Principles

Distribution beats features.

We chose WhatsApp over an app. Users were already there.

Fix the funnel before the model.

Our biggest conversion jump came from fixing scheduling, not the model.

Evals are the spec.

If you can't score it, you can't ship it.

Mistakes are tuition.

I've paid plenty. Ask me about my favourite one.

Say hi

Say hi. Dobby will make sure I see it.

Building agents, fixing a leaky funnel, or want to argue about evals? I'm in. I reply faster than most agents I've built. Dobby disputes this.

LinkedIn ↗ GitHub ↗

Notes you add to the board go too. Send as email
/