Phase v0.1 — schema and confirmation (Q1 2027)

Published intent schema v1 covering book, pay, query, cancel across 4 verticals.
Numeric confirmation flow (press-a-digit).
ASR + TTS in 4 launch languages: English (UK + Indian), Hindi, Swahili, Armenian.
Reference client: integrates with Twilio and LiveKit runtimes.
First production integrations: GeraClinic, GeraEats.

Open question: the latency budget on 2G. If the 1.2-second end-to-end target slips, the feature-phone flows need rework.

Phase v0.5 — fallback SLA and vault integration (Q3 2027)

Human-fallback SLA: median handoff under 8 seconds, p95 under 20.
10 additional languages: Yoruba, Hausa, Amharic, Urdu, Bengali, Georgian, Arabic (MSA + Egyptian), Portuguese (BR), Spanish (LatAm), Russian.
GeraMind vault integration for preferences and consent.
Audit-log export for operators.
Published per-language word-error-rate dashboards.

Open question: human-fallback unit economics at scale. If cost per escalation exceeds a threshold, we narrow the refusal triggers.

Phase v1.0 — coverage and enterprise (late 2027)

100+ languages with varying depth (core 30, extended 70).
Code-switching support (bilingual ASR for the 12 most common pairs).
Enterprise onboarding with multi-tenant isolation and private-branch deployment.
Integration with GeraNexus for signed transaction receipts.
v1.0 spec freeze with 2-year stability commitment.

Open question: low-resource-language ASR quality. Depends partly on external model investment.

What we will not do in v1

Voice biometrics as a primary auth factor.
Advertising / upsell scripts inside transaction flows.
Call-recording training on user voices without explicit, per-call, opt-in consent.

How to follow

Draft spec: geravoice.com/spec. GitHub issues for debate. If you work on ASR, low-resource languages, or voice-agent runtimes and want eyes on v0.1 before freeze, email [email protected].