Product applications
Web and native mobile apps, from first prototype to the App Store and Google Play. React, Next.js and TypeScript, shipped on infrastructure you own and can hand to another team if you ever need to.
We build AI apps
that hold up to scrutiny.
Riff Apps is a UK studio. We take an idea from first sketch to a product people use every day — with data protection, accessibility and AI governance designed in from the first sprint, not bolted on the week before launch.
- Riff Social app, live on iOS and Android
- TeachWise AI Personalised curricula, live on the web
- Project controls In build for the energy sector
What we build
Five things, done properly, rather than a list of everything. Most engagements combine two or three.
AI features that earn their place
Conversational assistants, generated content, matching and classification — built on current frontier models, with evaluation, guardrails and a human path for anything consequential.
Data and integrations
Schema design, migrations, and connections into the systems your business already runs on.
Governance and assurance
DPIAs, records of processing, model documentation and accessibility audits, produced as the work happens.
Run and improve
Monitoring, incident response, security patching and a steady release cadence after launch.
Selected work
Two products live, one in build. Each one shipped end to end by the same team.
Riff
A social app that connects people through questions, voice and trust instead of photos and endless scrolling. Twenty‑five questions across values, goals and communication style feed a compatibility model; faces are revealed only when both people are ready.
It carries the harder parts too — three‑tier ID verification, liveness checks, encrypted voice and media, and a layered safety system watching for harmful behaviour.
Visit riff-app.co.uk Compatibility 4‑layer score
Voice note 0:42
Verification Green
Reveal Both ready
TeachWise AI
A learning platform that builds a curriculum for whatever someone needs to learn next. Rather than handing over a fixed course catalogue, it assembles a structured path around a learner’s goal and moves with them as they progress.
The interesting engineering is in keeping generated material coherent across a whole syllabus — sequencing, prerequisites and consistency between lessons, not just good output one prompt at a time.
Visit teachwise-ai.com Goal Set
Curriculum 9 modules
Progress
Next lesson Ready
Coming next: project controls for energy
Our current build is a project controls application for the energy sector, with the same approach heading into adjacent industries.
Capital projects in energy run on cost, schedule, change and risk data that usually lives in a dozen spreadsheets and one very tired planner’s head. We’re building a single application that holds the baseline, tracks progress against it, and explains variance in plain language.
AI does the reading — parsing progress reports, flagging drift, drafting the narrative for a monthly review — while every number stays traceable to its source and every forecast stays something a human signs off. The same core suits construction, utilities, infrastructure and manufacturing, and we’re talking to teams in each.
Talk to us about an early pilot Cost performance CPI 0.97
Schedule float 4 weeks
Open changes 12
Forecast Signed off
How a project runs
Short cycles, working software early, and the compliance paperwork produced alongside the build rather than after it.
Frame the problem
A week or two of discovery: who the users are, what decision or task the product has to improve, what data exists, and what the regulatory picture looks like. You get a scope, a budget and a risk register — and an honest answer if we think the idea needs changing.
Prototype something clickable
A working prototype in front of real users before the architecture is locked in. It is far cheaper to discover a flawed assumption in a prototype than in production.
Build in two‑week cycles
Every cycle ends with something deployed to a staging environment you can use. Weekly written updates, a live board, and no surprises at the end of the month.
Assure before launch
Accessibility testing against WCAG 2.2 AA, a security review, penetration testing where the risk warrants it, a data protection impact assessment, and model evaluations for any AI feature that affects users.
Run it, then improve it
Monitoring, alerting, dependency patching and a regular release rhythm. Most of our work is with clients we launched for a year or more ago.
Governance is part of the build
Anyone can add an AI feature. The work is making it defensible to a regulator, a board and the person on the other end of it.
People know when they’re talking to AI
Generated content and AI participants are labelled in the interface, every time. We don’t build products that pass off a model as a person.
Consequential decisions keep a human
Where an output affects someone’s money, safety, learning or access to a service, a person reviews it and can overturn it. That route is designed in, not promised in a policy.
Data minimised by default
We collect what a feature needs, keep it for a stated period, and document the lawful basis before we write the code.
Evaluated, not assumed
AI features ship with test sets, measured failure modes and monitoring, so quality is a number rather than a feeling.
Accessible as standard
WCAG 2.2 AA is the baseline for everything we deliver, tested with a keyboard and a screen reader.
Tell us what you’re trying to build
An idea on the back of an envelope, a prototype that needs finishing, or a platform that needs to pass an audit — we’ll tell you honestly whether we’re the right team for it.