Problem. Approach. Outcome.
Short stories about real problems we've solved – in our own products and for clients. Each one shows what we did and what changed.
Real problems we've solved for clients
Rescuing a product that was falling apart under load
A growing SaaS company's application was crashing during peak hours. Their small dev team had moved fast to ship features but accumulated significant technical debt. Response times were climbing and customers were complaining.
- Conducted a full technical audit to identify the root causes – not just symptoms
- Prioritised fixes by impact: database queries, caching, and error handling first
- Refactored the three worst-performing endpoints and added proper monitoring
- Documented the architecture and created runbooks for the team to maintain it
The same approach, applied to our own bets
We use the same playbook on our in-house products as we do on client builds. Every pattern we recommend has been tested in production first.
Making self-assessment tax filing simple
UK self-employed workers dread tax returns. The process is confusing, the language is impenetrable, and mistakes lead to penalties. Most pay an accountant hundreds of pounds to avoid the stress.
- Mapped every HMRC self-assessment journey into plain-English, step-by-step flows
- Built guided forms that ask one question at a time, with contextual help at every stage
- Integrated directly with HMRC APIs for real-time submission and status tracking
- Added automatic tax calculations so users see their bill update as they go
Building AI learning tools children can use safely
Children want to use AI for homework help, but existing tools aren't designed for young users. Parents worry about inappropriate content, data privacy, and unsupervised screen time.
- Designed a child-friendly AI interface with age-appropriate language and guardrails
- Built a parental control layer that gives parents visibility and control over all activity
- Created adaptive quizzes that adjust difficulty based on the child's responses
- Implemented content filtering and safety boundaries at every interaction point
Matching people to jobs they'll actually enjoy
Traditional job boards rely on keyword matching. Candidates apply for dozens of roles they're not suited for. Employers waste time filtering irrelevant applications. Nobody's happy.
- Built an explainable AI scoring engine that matches on capability, not just keywords
- Designed clear match explanations so candidates understand why a role suits them
- Created employer dashboards that surface the best-fit candidates first
- Added skills-gap analysis that shows candidates what they need to develop
From idea to build-ready in minutes, not months
Founders and solo developers spend weeks scoping, writing specs, and figuring out where to start. By the time they're ready to build, momentum has stalled and budgets have shrunk. There was no structured way to go from a raw idea to actionable build instructions without expensive consultants or lengthy discovery phases.
- Built an AI-powered pipeline where multiple models independently review and score each idea, surfacing strengths, risks, and feasibility before committing to a build
- Enforced a structured pipeline – each stage must complete before the next begins, preventing incomplete specs from reaching development
- Generated IDE-specific build instructions tailored for Replit, Cursor, Lovable, Warp, and other popular environments
- Added execution safety controls that classify every build step for risk and flag irreversible actions like database migrations before they run