Menu
Case file 01 · Marketplace · India

Karkaana

Tinder for factories in India. Type what you want to make, swipe through real manufacturers, send a quote request.

File summary

Pre-launch
Filed under
Marketplaces
Built by
Varun Chennuri, with an AI engineering team

01

The problem

A founder who wants to make a product, say protein chips, first has to find a manufacturer that can actually make it.

In the US, Keychain built a platform for exactly that search. Before writing any code, I reverse-engineered Keychain, then built the India version on factories that exist on the public record.

02

What I built

Karkaana works like Tinder for factories.

  • Type what you want to make ("protein chips") and see real manufacturers.
  • Swipe to shortlist, then send a quote request.
  • 94,000+ real MCA-registered factories, indexed from the public company master.
  • The database itself enforces that only public-record fields are shown.
  • Login with a WhatsApp code, payments through Razorpay, and a ₹10 GST verification step.
  • An open-source website and footprint finder.
  • Codex-generated category art, and a mascot named Dabba.

03

How AI agents built it

Reverse-engineer first, then build with the two-model team.

  1. I took apart Keychain, the best product in the category, before building my own.
  2. Built with my AI engineering team: Claude Code and Codex build in parallel and review each other's code, then tests, then a pull request.
  3. Codex also generated the category art.
  4. 86 commits in about 17 days, starting 22 Oct 2026.
See the full method

04

Architecture sketch

Drawn the way it sits on my desk: boxes for the moving parts, arrows for the flow, dashed lines for what is wired in alongside.

  1. 01FOUNDERtypes "protein chips"
  2. 02SEARCH + SWIPE APPNext.js
  3. 03FACTORY INDEX94,000+ MCA-registered factories Neon Postgres only public-record fields shown, enforced in the database
  4. 04SWIPE TO SHORTLIST
  5. 05QUOTE REQUEST
  6. 06LOGIN + PAYWhatsApp code Razorpay ₹10 GST check
  7. 07EXTRASfootprint finder (open source) Codex art mascot: Dabba
Karkaana architecture sketchA founder types what they want to make into the Next.js search and swipe app. It queries a Neon Postgres index of 94,000+ MCA-registered factories that only exposes public-record fields, enforced in the database. The founder swipes to shortlist real manufacturers and sends a quote request. WhatsApp-code login, Razorpay and a ₹10 GST check are wired in, along with an open-source footprint finder, Codex category art and the mascot Dabba.FOUNDERtypes "protein chips"SEARCH + SWIPE APPNext.jsFACTORY INDEX94,000+ MCA-registered factoriesNeon Postgresonly public-record fields shown,enforced in the databaseSWIPE TO SHORTLISTQUOTE REQUESTLOGIN + PAYWhatsApp codeRazorpay₹10 GST checkEXTRASfootprint finder(open source)Codex artmascot: DabbaQUERYREAL MANUFACTURERSALSO WIRED IN
Karkaana architecture sketch:A founder types what they want to make into the Next.js search and swipe app. It queries a Neon Postgres index of 94,000+ MCA-registered factories that only exposes public-record fields, enforced in the database. The founder swipes to shortlist real manufacturers and sends a quote request. WhatsApp-code login, Razorpay and a ₹10 GST check are wired in, along with an open-source footprint finder, Codex category art and the mascot Dabba.

05

Results

Real numbers only. Anything not confirmed yet is marked as a placeholder.

  • 94,000+real MCA-registered factories indexedBasis: Public company master
  • 86commitsBasis: Commit history
  • ~17 daysof building, from 22 Oct 2026Basis: Commit history

06

Stack

  • Next.js
  • Neon Postgres
  • WhatsApp-code login
  • Razorpay
  • GST verification
  • Open-source footprint finder
  • Codex
StatusPre-launch

Next step

Want this for your company?

Start a 7-day sprint and get a working AI product: two AI models that review each other, tests on every change, and me on every risky call.

Start a sprintRead how I build