Portfolio · AI Product Build

MobiX

Mobile app design and development for a community platform — from a stalled build to a shipped, agentic-assisted release.

The client

A stalled mobile build.

MobiX had a mobile app in development for nine months with an offshore team, a product spec that had drifted from what was actually being built, and no clear release date. The founders needed a build partner who could both finish the product and tell them honestly whether the original spec still made sense.

The challenge

Ship without inheriting the drift.

Taking over a nine-month-old codebase meant deciding, fast, what to keep, what to rebuild and what to cut — with a founding team anxious about sunk cost. The technical debt was real, but so was the pressure to protect what had already shipped to a small existing user base.

A second challenge was scope: the original spec had grown to cover four distinct user journeys, only one of which had any usage data behind it.

The approach

Cut first, then build.

01A one-week auditCodebase, spec and usage data reviewed together before any commitment on what to keep.
02Cut to the proven journeyThree of four planned user journeys shelved in favour of the one with actual usage signal.
03AI-assisted where it earned itContent moderation and onboarding personalisation built agentic; everything else kept deliberately simple.
Delivery

From audit to release.

01
Codebase and spec audit
What to keep, what to rebuild, what to cut — decided against usage data, not opinion.
Week 1
02
Scope cut and rebuild plan
Three unproven journeys shelved; the core flow rebuilt on a stable architecture.
Weeks 2–3
03
Core build and AI modules
Moderation and onboarding personalisation shipped as agentic modules with their own evals.
Weeks 4–11
04
Release
Shipped to existing users first, then opened to new acquisition.
Week 12
What shipped

A finished product, not a rescue story.

01Rebuilt core appThe one proven user journey, shipped on stable architecture.
02Agentic moderationCommunity content moderation with human escalation and an evaluation set.
03Onboarding personalisationAI-assisted onboarding tuned against early retention data.
04Deployment and handoverCI/CD, monitoring and documentation handed to the in-house team.
The core flow
The one user journey with real usage signal, rebuilt on stable architecture.
Moderation module
Agentic content moderation with human escalation built in.
Onboarding pass
Personalised onboarding tuned against early retention data.

Yes — most builds are not greenfield. The audit exists specifically to make that decision on evidence rather than sunk-cost pressure.

Against usage data where it exists, and against the venture thesis where it does not. Ambition is not the same as evidence.

Scoped to requirement

We operate, we do not observe.

We build alongside the founding team and hand over something they can run without us.

Talk to the studio