AI Development Case Studies

Both — depending on what the project needs. For HeyPractice, we researched and integrated third-party AI services (IBM Watson for speech-to-text, PRECIRE for psychological text analysis) after evaluating several providers. For EveryPig, we connected a specialist ML module to a platform we built. For AreaButler and Lake, we integrated OpenAI's GPT models directly. The right call depends on your use case, timeline, and budget — we'll recommend the approach that ships fastest without creating technical debt.

Yes, and most of these case studies are examples of exactly that. AreaButler already had an MVP when they came to us — we augmented their team and integrated GPT-3.5 to generate location texts automatically, which made them the first PropTech in the DACH market to offer that feature. Lake was already live when we joined — we rebuilt the platform and added ChatGPT for property description generation. You don't need to start from scratch.

Across these projects: AI speech recognition and audio-to-text transcription (HeyPractice), emotional tone and keyword analysis from video recordings (HeyPractice), disease and mortality prediction from farm health data (EveryPig), GPT-powered location and property description generation (AreaButler, Lake), and proximity-based smart search (Lake). The common thread is AI that reduces manual work or gives users instant feedback they couldn't get otherwise.

EdTech (HeyPractice — AI sales training platform, Germany), AgriTech (EveryPig — swine health management platform, USA), PropTech (AreaButler — AI real estate tools, Germany), and Travel (Lake — vacation platform with smart search and content generation, USA). We're not limited to these — the AI patterns repeat across industries.

We treat it like any engineering decision: we research, test, and switch if needed. On HeyPractice, we integrated AssemblyAI first — it turned out not to support German at launch despite promises. We pivoted to IBM Watson mid-project and shipped on schedule. That kind of research and contingency work is part of what we deliver, not a billable extra.

We work in 2-week sprints with demos at the end of each. For AI features specifically, there's usually a research phase where we evaluate providers or architectures before writing integration code — this is where problems get caught early. Most projects start with 3–8 people depending on scope. AreaButler ran lean with one full-stack developer embedded in their team. HeyPractice scaled to 8 people across frontend, backend, and QA. We size the team to the problem, not the other way around.