The current prototype
What works, measured latency and technical constraints of the current prototype.
Read the prototype’s actual state
What is connected today, what still needs proof, and where the next test sessions should focus.
A working state, not a promise
Feasibility is established; the public experience still needs to be proven.
The prototype already connects conversation, memory, orchestration, images and observation for Max and Emma. An available setting or a technical trace does not replace human evaluation of the encounter. Listening, coherence and comfort in real conditions remain decisive; the Roadmap presents hosting and presentation conditions.
Feasibility established
Working parts can run together and be observed.
Experience to prove
Presence, pacing and public conditions still need testing.
Prototype capabilities
What already works together
A voice conversation linked to the journey
The interface, speech capture, transcription, generation and synthesis can already form one conversational loop.
- Web interface and journey
- Voice conversation, transcription, generation and synthesis
- The journey and listening conditions are described in Experience
Tap a block to read its detail.
The tests that matter
What still needs to hold
Does artistic quality hold over time?
A functioning loop does not guarantee that a participant feels genuinely in relation with a character, nor that pacing stays right across several exchanges.
- Artistic quality over time
- Relevance and dosage of cinematics
- Feedback expressed by participants
Tap a block to read its detail.
The Game Master’s role
Direct a relationship, not every line
- 1
Intent
An invisible director protects pace, trust and the moment for an image.
- 2
Implemented
It acts after the speech turn; it does not block each answer.
- 3
To assess through listening
A video choice waits until speech ends. Its timing and effect are assessed by listening to the whole conversation.
Latency — narrative challenge
Design budgets and technical measurements are kept separate: neither replaces listening to the conversation.
The measurements below describe the prototype’s technical behaviour. They do not replace human listening or promise public performance. Latency is also a staging concern: a pause may be perceived as hesitation, emotion or a malfunction. The diagram before the PostHog panel is a design budget, not an achieved performance. The Game Master acts after the speech turn without blocking the voice response.
Latency budget — orders of magnitude
Cumulative historical budgets (excl. STT, excl. avatar, incl. pre-turn GM): ~2.8 s. Experience design target: ~2 s — not an achieved performance. Minimal voice cascade market reference: ~500 ms in the best documented case, not a prototype measurement.
Sources: Notion Où est Ava ? pipeline (prototypeDiagrams), portal benchmarks (sttData, pipelineData, WhatWeHaveDiagram).
Browse each link. Design budgets and market references are not prototype measurements. The pre-turn Game Master is a historical step, off the critical path. Technical measurements in the PostHog panel can be compared with human session observations.
Streaming voice transcription via WebSocket + VAD — first measurable link in an audio turn.
Services and role
- →Où est Ava ? orchestrator
- →Deepgram (transcription)
- →Interchangeable STT services
Range / budget
Market: ~75–200 ms (Deepgram Nova, streaming P90 / typical)
Narrative stake
Slow transcription delays the entire pipeline; the user already hears silence before the character even 'thinks'.
Tap an item to show its detail below.
Browse each link. Design budgets and market references are not prototype measurements. The pre-turn Game Master is a historical step, off the critical path. Technical measurements in the PostHog panel can be compared with human session observations.
Streaming voice transcription via WebSocket + VAD — first measurable link in an audio turn.
Services and role
- →Où est Ava ? orchestrator
- →Deepgram (transcription)
- →Interchangeable STT services
Range / budget
Market: ~75–200 ms (Deepgram Nova, streaming P90 / typical)
Narrative stake
Slow transcription delays the entire pipeline; the user already hears silence before the character even 'thinks'.
Analytics & sessions
PostHog metrics — latency, journey and technical session traces.
PostHog technical monitoring
Monitoring page to diagnose waiting times and compare them with human observations of sessions, without automatically inferring experience quality.