Book a strategy call
Back to case studies
Education Administration Programs, AI

Speak Easy: Creating an AI Voice Translation Product for Universities

How a Spanish-British education startup engaged a Solution Architect, a Business Analyst, a UI/UX Designer, a Frontend Engineer, a Backend Engineer, a DevOps Engineer, a combined Project Manager and QA Engineer, a Streamlining Consultant and a Low-Code Engineer.

Challenge

Speak Easy is building real-time voice translation into video calls for universities: someone speaks in one language during a seminar and the people listening hear their own. Two things make that hard at once. The product is held to numeric accuracy thresholds rather than impressions, because a translation that is approximately right in a lecture is worse than no translation at all, and the quality has to hold while speech is being transcribed, translated and spoken back inside a live call. At the same time it is an enterprise integration problem: the application lives inside Microsoft Teams, runs on Azure, authenticates through institutional single sign-on, and sits behind an admin panel where universities manage organisations, users and subscriptions. Behind all of it was a solo non-technical founder - no CTO, no engineer, and nobody in the company able to write the specification or judge the answers.

That is why the engagement began with definition rather than development. A solution architect with six and a half years across systems architecture, database design and infrastructure, working across the JVM stack with Spring, Micronaut and Play, and comfortable in distributed international teams. A business analyst with two decades of turning how an organisation works into requirements precise enough to build from. And a project manager who also ran quality assurance - five years of manual and API testing across web, mobile and desktop, functional, regression and integration testing against microservices and databases, combined with the delivery side: requirements turned into epics and user stories, and agile practice adapted to the team rather than imposed on it.

The build team followed. A UI/UX designer, five years across fintech, banking and SaaS, working in design systems, research and prototyping. A frontend engineer with eleven years of JavaScript behind them, single-page and progressive web applications, performance work, and enough delivery experience to lead. A backend engineer, six years in Python frameworks, used to owning complex server-side processes. A DevOps engineer with five years on AWS infrastructure, Kubernetes, infrastructure as code and CI/CD. Alongside them, two specialists: a streamlining consultant to keep a nine-person distributed team working as one, and a low-code engineer for the internal and administrative surfaces that did not warrant bespoke development.

9
Talents Secured
8+
Months Duration
48h
Time-to-hire

Outcome

A founder without a technical background cannot manage nine technical people. The engagement was structured so that the architecture, the requirements and the delivery management came first and the rest of the team worked to those, which meant the founder was answering questions about the product rather than about the engineering. Nine roles, one request at a time, over eight months.

  • Nine roles across definition, design, build and specialist support, with the first engagement confirmed in 48 hours.
  • Requirements, architecture, design, frontend, backend, infrastructure, testing and delivery management - an entire product organisation for a single product.
  • Assembled around a solo non-technical founder with no in-house engineering of any kind.
  • Eight months covering specification, design and build of a translation application inside Microsoft Teams, with its administrative platform alongside it.

A Strategic, Multi-Phase Hiring Approach

Phase: Request

Nine requests, opened one by one: architecture, analysis and delivery management first, then design, frontend, backend and infrastructure, with a streamlining consultant and a low-code engineer added as the product’s surface demanded.

Phase: Matching

With no technical counterpart inside the company, everyone out of 27 shortlisted had to be self-directing and able to explain their reasoning to a non-technical founder. Beyond that, each request was screened on its own ground: architecture owned through to production, requirements that had survived being built from, and testing against microservice systems rather than screens alone.

Phase: Interviews

The founder interviewed and decided throughout, with the first engagement agreed 48 hours after the opening request.

Phase: Engagement

Nine engagements opened with a mix of full-time and part-time monthly workload, daily hour tracking, and monthly reporting with detailed breakdown of hour commitments.

Phase: Support

Across eight months as the work shifted from definition into design and then into build all changes to workloads were requested and approved from the account.