case study / live MVP

myberuf.com

A German-for-work learning product built around workplace situations, AI coaching, mistake review, progress tracking, and handover-heavy AI-agent development.

live MVPMistral workflowbrowser-local progress

What this case study covers

$ case-study --scope honest

Voice automation experiment

$ voice-practice --latency-matters

myberuf's long-term direction is not only static typing. Workplace practice can become spoken, interactive, and simulation-like: the learner should eventually be able to speak through interview, onboarding, feedback, and Rückfragen moments.

A Vapi / voice automation experiment explored what it might feel like for learners to speak to the platform instead of only typing. It was an experiment, not a fully shipped feature, and it made one product lesson very clear: voice changes the AI stack.

$ source --voice-experiment Vapi voice experiment

Source video from the myberuf voice/automation experiment. The public page links to Loom instead of relying on a local iframe player that can fail during preview.

Typed coaching can tolerate a short delay. Spoken workplace simulation feels broken if the response takes too long. In voice learning, latency becomes part of the learning experience.

That changes the model-selection question from "which model is smartest?" to "which model is good enough, fast enough, affordable enough, and trustworthy enough for this interaction?" For a European German-for-work product, Mistral is strategically interesting because model-layer choices affect latency, trust, deployment fit, cost, and privacy expectations.

Status

This is a living case study. It can be expanded with more screenshots, source-backed build notes, and private-beta results later, without exposing private implementation details or secrets.