Voice AI · Bank collectionsPrototype · Award winner
The problem: banks chase early arrears with scripted call centres that miss the customer's language, improvise on payment terms, and leave a thin record of what was actually agreed.
The thesis: a governed outbound voice agent that speaks Emirati Arabic, Emirati English and Indian English, verifies identity, offers only the payment plans a decision engine has approved, and logs every commitment. Built with Team Mizan on ElevenLabs Agents, it won the Dubai AI Voice Challenge at the Dubai AI Festival on 27 October 2026.
Creator intelligencePrototype
The problem: brands pick creators on follower counts and instinct, then hand them briefs full of claims the product cannot back.
The thesis: mine real audience signal from TikTok and Instagram, turn it into an evidence-backed, claim-safe creator brief, and draft the ad with an AI presenter. Built in September 2026 with FinalTake.ai; the working demo is live.
MicromobilityPrototype
The problem: short city trips meant haggling for a ride or owning a two-wheeler you barely used, and the rentals that existed made you return the bike where you found it.
The thesis: one-way, pay-per-kilometre rentals: pick up at any station, drop at any other, unlock with a QR code. It was the founder's first startup, in Jaipur in 2018, and it did not survive; the lessons did. The app prototype and original deck are public while the relaunch is replanned for Dubai.
Live sports engagementPrototype
The problem: watching live sport on a second screen is passive. The momentum everyone feels in a match has no visual, interactive form.
The thesis: render match momentum as a living probability curve, with a locked design foundation that keeps the product's identity stable no matter which teams play. Interface prototypes and the full design system exist; the engine decides next whether it graduates to build.
NetworkingPrototype
The problem: after a good conversation at an event, the contact ends up as a crumpled card or a name you half remember.
The thesis: the other person says their details out loud, the app captures them with a selfie, and the follow-up goes out on WhatsApp before you leave the room. Built at the Collabute x TheBlock hackathon in Dubai, August 2026.
AI governancePrototype
The problem: governments and regulated firms want to run AI on servers they own, but cannot prove afterwards what a model was asked and what it answered.
The thesis: a policy proxy in front of the model that allows, flags or blocks every call, then seals it into a signed, hash-chained record an independent verifier can check. The live demo catches a record forged outside the app.
Personal voice AIInternal R&D
The problem: a founder running several products spends the day hopping between chats, pull requests and calendars just to find out what needs them.
The thesis: NoNu, a voice assistant that phones the founder, knows every project, hands over to a lead persona for each product, and acts on WhatsApp or email only after a spoken secret word. Private, and in daily use.
Engineering infrastructureInternal R&D
The problem: a small team cannot ship venture-grade software on multiple fronts with conventional processes. Speed usually costs discipline.
The thesis: coordinate specialised AI engineering agents, each with a role, a doctrine of hard-won engineering lessons, and non-negotiable review gates, under a human founder's final pen. The Fleet is the harness that built and maintains Snook A Look in production; five Compose Multiplatform components it shipped are open source. It is infrastructure, not a product for sale, and it is the engine's biggest unfair advantage.
Concepts stay in the lab until they earn a status. When a venture is archived, it will say so here. What you see is the honest state of the engine.
