Engineering AI into software delivery: How Travelopia launched software to production

Before engaging Equal Experts, Travelopia had already spent more than a year exploring AI tooling. Motivated by a desire to improve developer productivity and gain a competitive advantage with AI-driven gains, Travelopia ran internal hackathons, tested tools like Cursor and GPT-3, and trialled code generation experiments. These activities surfaced several opportunities — but also highlighted the risks of adopting GenAI without structure.

Early concerns included:

• IP and data security — how to keep control of proprietary knowledge while using powerful LLMs
• Code quality and maintainability — AI-generated code varied in reliability, raising doubts about long-term sustainability
• Governance and repeatability — teams were working without a clear framework to guide adoption

Travelopia needed a way to make sense of their early AI experiments and begin identifying opportunities that could be scaled. Rather than focusing on broad transformation or one-off demos, they wanted to explore how AI could be integrated responsibly into a real production system — one that could showcase tangible business value while laying a foundation for repeatability. They partnered with Equal Experts to identify an ‘exemplar opportunity’ — a focused initiative that, if executed successfully and shown to deliver meaningful results, could serve as a compelling proof point for how AI might be scaled across the organisation.

For our exemplar opportunity, the teams selected a legacy system rewrite as an ideal candidate for GenAI adoption — specifically, a platform to score incoming leads and assign them to the right consultant.

Our approach is designed to deliver gains and time to value faster, without piling up complexity and unmanageable technical debt — whilst keeping speed, quality and maintainability at the core. That’s why we use our three-phased approach to AI-powered software delivery:

• Phase 1: Augmented expertise — embedding GenAI into existing workflows to improve productivity without major structural changes
• Phase 2: Accelerated initiatives — designing small, focused teams around GenAI-first delivery practices to unlock strategic value
• Phase 3: AI organisations — rethinking team structures, delivery models, and governance to fully integrate AI into the operating model

With Travelopia, we saw an opportunity to move beyond Phase 1 experimentation and focus on Phase 2: building a real, production-grade AI-powered initiative. Our goal was not just to generate code faster, but to design and deliver something valuable, repeatable, and well-engineered – but with reduced time-to-value and cost-per-feature.

We selected a legacy system rewrite as an ideal candidate for GenAI adoption — specifically, a platform to score incoming leads and assign them to the right consultant. We designed and built a back-end-first system with a user interface for sales managers to manage consultants and oversee lead lifecycles, with integrations to existing CRM and communication tools. The engagement focused on hands-on delivery while embedding the practices that would support long-term GenAI adoption.

Engineering rigour meets AI enablement

Despite generating the vast majority of implementation code with LLMs, we rejected the trend of “vibe coding.” Instead, we applied:

• Human-driven architecture: All system boundaries, service decomposition, and data models were designed by the team, not the AI
• Structured decomposition: Responsibilities were cleanly separated across microservices and Lambda functions
• Clear standards and examples: Reusable code snippets and language-specific rules guided AI behaviour
• Mandatory code reviews: AI-generated outputs were always reviewed by developers for consistency and quality
• Prompt engineering discipline: We created detailed, example-led prompts that produced reliable, context-aware code

By treating AI as a “design and implementation partner” — not an autonomous implementer— the team maintained full control over the system’s design integrity.

Foundation-first, UI-second

While stakeholder interest naturally gravitated toward the user interface, we prioritised building the system’s core first, that consisted of:

• A robust lead scoring algorithm
• A flexible consultant matching service
• A secure, validated data layer
• A state machine that orchestrated, and audited, lead flow
• An API layer designed to support multiple front-end paths

This allowed the UI to evolve rapidly once the core system was in place — a classic payoff of strong architectural foundations.