After a few weeks of stabilizing our direction, Week 9 was about deep refinement — both in technology and experience design. Our main focus was to test and upgrade our conversation system while clarifying how the user experience should be structured for the upcoming playtest.
Experimenting with ElevenLabs and ChatGPT 5
As planned after the half presentation, we switched our tech stack from ChatGPT-4o to ChatGPT 5 plus ElevenLabs.
The motivation was simple: we needed more control and better realism.
With this upgrade, we were finally able to:
- Build a system that allows discrete control over conversation flow rather than relying on one continuous stream of output.
- Generate more natural responses that sound closer to real, three-way human conversations.
- Create distinct and imperfect voices — with tone shifts, pacing, and even small verbal tics — that make Athena and Poseidon feel alive rather than robotic.
Compared to 4o, the difference was dramatic. GPT-4o was a compromise: it could handle both voice input and voice output, but neither well. The logic layer was limited, the voice quality synthetic. GPT 5 and ElevenLabs combined gave us the best of both worlds — a smarter brain and a more expressive voice

Redefining the Experience Design
While the tech improved, we also needed a clearer structure for our conversations. Over several design discussions, we formalized three distinct phases of interaction:
- Open Talk – Users can freely ask any question they want.
- Scripted Discussion – Pre-written exchanges that ensure specific experience goals are met.
- Generated Discussion – Longer, AI-driven back-and-forth dialogue where Athena and Poseidon continue a topic naturally across several turns.
By classifying conversations this way, we gained a higher-level control of the player’s overall journey. Instead of the outcome depending entirely on what users ask, we can now deliver consistent experiences — ensuring that each of the 100 students playing the demo will still feel a similar narrative arc.
Looking Ahead
By the end of Week 9, we had a much clearer understanding of both our technical foundation and experience design framework.
Next week will be all about integration — making Athena and Poseidon’s voices sound even more natural, tuning the dialogue flow, and preparing a stable build for Week 10’s playtest day.
