The Challenge
This is what happens when code stops being the bottleneck
The Challenge
Happy Student is a UK EdTech startup on a mission to improve student wellbeing. They cinnect university students with mental health support, local deals, and resources through a mobile app.
After gaining strong traction in the UK, Founder and CEO Aram Tufan had secured a pivotal agreement to launch at Georgetown University in Washington DC on 9th March 2026. A critical date timed to coincide with students returning from spring break, and a launchpad into a network of elite East Coast universities.
There was just one problem: the app was built entirely for the UK market. And it hand... problems.
When Dootrix first engaged the project in late January 2026, the situation revealed further complexities:
-
No documentation. The app had been built over 18 months by a previous development team, but there was no technical specification and no documented understanding of what existed, how it worked, or how it was structured.
-
Inherited codebase. Dootrix had not written the application. The codebase required thorough assessment before a single line of new code could be written responsibly.
-
Critical bugs. Initial testing showed the app had repeatable cashes in production.
-
Zero test coverage. The backend had no unit tests whatsoever.
- Security gaps. A number of security issues and legacy assets that required hardening.
-
No US functionality. There was no country/state-level content system, no US-appropriate deals or other content, no US Voiceflow chatbot configuration. The app simply did not know America existed.
- Extreme time pressure. From contract signature to App Store launch: just over six weeks.
The Georgetown launch was not an arbitrary milestone. They had agreed to act as a liaison to other universities across the DC and Boston areas. A successful launch could open the entire US higher education market.
"Thank you so much for all the fixes and improvements you've made. Quite frankly, this process has shown us there are no limits to what we can achieve together. The reactions from GU students, academics, and the administration team were all very positive. The US market is very responsive, and they keep saying we've created a great product!"
- Aram Tufan, CEO, Happy Student
(on the Georgetown University launch, March 2026)
The Approach
Phase 1: Understand Before Building
Before any development could begin, Dootrix needed to understand what they were working with. Rather than treating this as a purely manual exercise, the team deployed AI to dramatically accelerate and deepen the assessment process.
Using a combination of Claude Opus and OpenAI's Codex, the team synthesised codebase analysis with structured interview transcripts; a technical walkthrough from the previous development team, and a product walkthrough from the Happy Student team. This approach generated a comprehensive technical assessment report in a fraction of the time the same exercise would traditionally take. What had previously been a 10-day exercise could, with this AI-augmented approach, be completed in a single day once source materials were in hand.
The assessment covered:
- Architecture and project structure
- Dependency health, third-party library risks, and deprecation timelines
- Security analysis
- Code quality, static analysis, and technical debt
- Infrastructure and deployment configuration
- Regulatory readiness (GDPR, App Store compliance)
- Monitoring, telemetry, and alerting gaps
- Documentation completeness
The result was a structured, actionable technical report that gave both Dootrix and Happy Student a clear-eyed view of the application's health as well as the confidence to proceed.
Retrospective Specification: The As-Is OpenSpec
As no specification existed, the team used AI coding agents to retrospectively generate a complete machine-readable specification of the Happy Student application directly from the codebase. Using the OpenSpec framework every feature, user journey, and functional behaviour was documented and committed into the repository.
The result was a human-readable, AI-readable specification that could be used to reason over what the application did, drive future development, and serve as the foundation for rebuilding or extending the app using AI coding agents.
This was the foundation of everything that came next.
To-Be Specification and US MVP Scope
With the as-is state fully documented, Dootrix then ran a focused specification phase led by a Senior Architect to define the US launch requirements. High-level requirements were translated into a structured technical specification using OpenSpec, with clear scope, dependencies, assumptions, risks, and MVP options. Everything was designed to drive specification-led development with AI coding agents.
Phase 2: Build at Speed
The defining characteristic of this phase was that virtually all code was written by AI coding agents; specifically Claude (Sonnet and Opus), and orchestrated by a senior Dootrix engineer.
Rather than writing code, the engineer's role shifted to: generating detailed, iterated proposals; reviewing and refining those proposals with agents cross-checking each other's work; and orchestrating execution against the plan.
Plans were developed iteratively through what the team called "manual Ralph loops"; a reference to the Ralph Wiggum Loop technique, where one agent would generate a plan or codebase change, another agent would critique it, revisions would be made, and the cycle would repeat until diminishing returns were reached.
This process front-loaded quality into planning, making execution close to "one-shot".
US MVP Feature Delivery
Dootrix delivered a comprehensive set of US-enabling changes across both the mobile app and the CMS. This included region specific hierarchies, US specific onboard and profiles, localised deal integration, US Voiceflow configuration, security remediations and the App Store releasex.
AI-Powered Content Generation for Universities
One of the most strategically significant additions was a new capability that allows universities to rapidly generate and deploy content assets into the Happy Student CMS using AI.
With brief, structured prompts, the system can generate relevant student wellbeing content including advice articles, support signposting, resource listings. This included a full draft/approval workflow before going live.
This dramatically reduces the time and effort required for a university to populate the platform with locally relevant, high-quality content, making each new institutional onboarding faster and more scalable.
UX Refinements
Alongside functional delivery, Dootrix completed targeted UX improvements to support the US launch including a refreshed home screen experience, improved onboarding flow clarity and tone, and an optimised App Store listing with updated visual assets and messaging.
ALL IN 6 WEEKS!
The Result
Happy Student launched at Georgetown University on March 2026. Right on schedule.