
Technical & pedagogical analysis
Edugami — Technical & pedagogical stakes
How GamiWays' conversational architecture serves active classroom pedagogy. Voice pipeline, pedagogical avatar, data sovereignty, emotional design.
15.10.2026 · Research — evaluation candidate
Inworld Inference
Proposed experiment, not integrated: compare the current LLM with Gemma 4 26B through Inworld on identical French exchanges. Measure documentary fidelity, unsupported claims, pedagogical adaptation, time to first audio and cost per completed turn; a lower price alone does not validate a model.
Published gains are vendor measurements, not results from our projects. Public inference is US-based; Enterprise regions and model availability require confirmation. ZDR must be activated and does not cover routed third-party providers.
01
Pedagogical positioning
The Edugami method
Edugami starts from a concrete dilemma — plastic pollution — to equip students to investigate, debate and structure their learning. AI is not presented as an automatic answer, but as one investigation tool among others, subject to critical thinking.
This approach follows flipped classroom pedagogy: the expert avatar prepares the ground, the interactive demonstrator engages students in the experience, and the teacher facilitates the collective debrief.
Link to GamiWays
Edugami is the first pedagogical deployment of the GamiWays pipeline. Where Où est Ava ? explores conversational cinema for adults, Edugami adapts the same architecture — expressive avatar, real-time voice, conversational memory — to a supervised school context.
School constraints (student data, GDPR, accessibility without installation) drive specific technical choices that in turn enrich GamiWays research.
02
Technical pipeline
Voice — STT / TTS
- —STT: Soniox v5 or Cartesia Ink-2 (WER < 2%, latency < 300 ms)
- —TTS: Cartesia Sonic 3.6 or Eleven v4 Turbo (vendor latencies, still to measure)
- —Edugami constraint: robustness across accents (French, English-speaking students)
- —No shared microphone: each student on their own device or collective demo mode
Pedagogical avatar
- —Real-time streaming video avatar (Simli Trinity-1 or Anam v3)
- —Adapted emotional expressiveness: calm and pedagogical, not dramatic
- —Target latency < 500 ms to maintain dialogue fluency in class
- —Edugami mode: short, sourced answers, active reformulation of student questions
Sovereignty & data
- —Data hosted in Switzerland / EU — GDPR and Swiss FADP compliance
- —No heavy installation: runs in the browser
- —No student account required for the basic experience
- —Anonymised student feedback used to improve the demonstrator
03
Emotional design — Edugami vs Où est Ava ?
| Dimension | Edugami | Où est Ava ? (Storygami) |
|---|---|---|
| Emotional register | Calm, pedagogical, encouraging | Dramatic, narrative, immersive |
| Avatar role | Expert guiding the investigation | Character living the story |
| Response length | Short, sourced, reformulated | Long, narrative, suspenseful |
| Conversational memory | Short session (90 min), reset between classes | Long session, narrative continuity |
| Audience | Students 12–18, supervised school context | Adults, festival / cultural context |
| Main constraint | GDPR, accessibility, multi-accent | Latency < 500 ms, max expressiveness |
Both prototypes share the same GamiWays pipeline but diverge on emotional design and operational constraints. Edugami constrains the system toward precision and pedagogy; Où est Ava ? pushes it toward expressiveness and immersion. These two poles mutually enrich the research.
04
Open questions
How far can we personalise without losing pedagogical coherence?
The demonstrator must adapt to the level and context of each class without drifting from the defined pedagogical framework. What granularity of personalisation is desirable?
How to measure real pedagogical impact?
Student feedback improves the experience, but how to assess critical skills acquisition beyond immediate engagement?
Scalability: from French-speaking Switzerland to the French-speaking world?
The offer is currently in French-speaking Switzerland and beyond. What are the obstacles to wider distribution — accreditations, cultural adaptations, institutional partnerships?
What role for the teacher in the conversational loop?
The expert avatar does not replace the teacher — it prepares and structures. How to design an interface that gives the teacher real-time control over the experience?