Senior Software Engineer

Three months on the backend of an AI voice product at Checkatrade, contracted through Evantis Technology. The initial contract ran to three months with an extension expected; a restructure meant it didn't go ahead, and the engagement closed at the original end date.

Getting Useful Quickly

The team had its own internal tooling and its own way of doing things — not the framework I'd have picked, not one I'd seen before. The first couple of weeks were mostly reading. That felt slow at the time and turned out to be the best investment of the engagement: most of the confusing behaviour I ran into later had an explanation sitting in the framework's source, and knowing where to look meant hours instead of days.

The other half of ramping up was the domain. AI-mediated conversations produce messy, ambiguous data, and a lot of the real work was deciding what a piece of that mess actually means before writing any code to store it.

Writing Everything Down

I kept a running knowledge base for the whole engagement — architecture notes, decision records, and dated investigation findings, written as the work happened rather than reconstructed afterwards. It started as a way to stop re-learning the same things and became the habit I'd most want to keep.

The payoff was constant rather than eventual. When a question came up in a meeting, the answer already existed and could be pointed at. When I picked work back up after a week on something else, the context was still there. When something behaved unexpectedly, there was usually a dated note explaining why it had been built that way. Nothing I figured out stayed only in my head, which is the difference between work that is finished and work that is merely delivered.

Key Contributions

• Feature delivery — shipped backend features across the product's API surface throughout the engagement, from data modelling through to production
• AI and voice integration — integrated AI voice capabilities into the product backend
• Config as code — moved AI assistant configuration and prompts out of vendor dashboards and into version-controlled repositories, so changes became reviewable
• Upstream contributions — contributed improvements back to the shared internal tooling rather than working around its limitations
• Compliance — delivered work covering auditable user consent and data erasure requirements, alongside the legal team
• Observability — treated alerting as something to be verified rather than assumed, and corrected monitors that looked healthy but could never have fired
• Documentation — every piece of work landed with its reasoning written down: architecture notes, decision records, and investigation findings, kept searchable for the team

Technical Stack

• Languages: TypeScript
• Runtime: Bun, Node.js
• Backend: OpenAPI-first services, REST, GraphQL, event-driven messaging
• Data: PostgreSQL
• Cloud: GCP, Kubernetes, Terraform
• AI: LLM integration, voice agents, prompt management
• Monitoring: Datadog, OpenTelemetry