We shipped the first Lina pilots on the GPT-4o transcription API and ripped it out three months later. Self-hosted Whisper large-v3 turned out faster, cheaper, and quieter on the DSGVO front, with the numbers to show it.
One webhook per call event collapsed the practice's n8n bridge during the 8 to 10 am peak. An append-only log with pull-based consumers fixed the retry storm in two and a half engineering days.
Voiceflow accelerated the first three demos, but the block model could not carry the uncertainty gate. What it took to replace it with a TypeScript node graph where the safeguards are middleware, not blocks.
Intent classifiers route to a handler; hand-off architectures route to a human. Why four parallel safeguards end up safer for medical calls, and the over-escalation trade-off you accept in exchange.
Signing the AVV is the easy 30 percent. The harder 70 percent is what the practice itself owes: patient notices, professional-secrecy authorisation under § 203 StGB, and the AI Act Art. 50 disclosure that becomes binding on 2 August 2026.