Ventures

From idea to working voice agent — in minutes, no code

Voice is the interface most businesses actually want and almost none can build. TalkifAI collapses that: describe the agent, connect a number, go live. I built the whole product surface — agent orchestration and telephony through to the dashboard, docs, and billing — provisioned for 20 concurrent live calls, around 400 an hour, with per-call analytics on duration, recordings, and transcripts. Every layer was instrumented: API, telephony, database, recording and egress, batch jobs, and RAG memory, with automated end-to-end call testing behind a public status page — 99.9% uptime, reported per component. It ran for SIEHS, the Sindh government emergency health service, and for HealthCloud, in 90+ languages.

Closed in 2026

Play-to-earn gaming for teens, on a revenue split

The model worked. Players earned around PKR 30,000 a month and we split it, so nobody needed money up front to start. Then the November 2022 crypto crash hit and token earnings fell roughly 99% — the same play that paid PKR 30,000 a month was suddenly worth about PKR 300. The unit economics never broke; the asset underneath them did. Players stopped logging in and I closed it rather than string them along.

Closed after the Nov 2022 crypto crash

A marketplace for custom accessories and printing

The bet was to take Pakistan Chowk — Karachi's dense, physical bazaar of small printers and accessory makers — online as a marketplace. Getting those sellers digital proved far harder than building the product, and it never reached the density a marketplace needs to work. It failed — but it taught me that marketplace liquidity, not software, is the real problem.

Shut down — never reached marketplace liquidity