Parallaxe is a conversational AI system for structured dialogue between people with opposing political positions. A value-based matching system brings participants together, while a real-time orchestration layer coordinates mediation, translation, clarification and source-based claim verification throughout the conversation.
I led the product and interaction design and developed the system architecture for the end-to-end AI-mediated dialogue experience. My work covered the multi-agent orchestration logic, conversational state and session management, multilingual processing, verification workflows, frontend implementation and the integration of the system into a deployable application.
Parallaxe
AI-mediated dialogue system
Cooperation with
Fraunhofer HHI
Pilot project
Berlin Science Week 2025
Featured in
Next gen AI mediation system.
Match
Pairs participants by maximal opinion distance using a structured intake.
Mediate
Reformulates and contextualises, translating across cultural context.
Verify
Flags verifiable claims and provides real-time, source-backed context where the evidentiary basis is sufficient.
Human – AI – Human
After being matched with someone holding substantially different views, participants speak to each other through a real-time mediation layer. Each contribution is transcribed and assessed before it reaches the other person: it can be translated, neutrally reformulated, followed up for clarification or checked against external sources. The participants remain in direct dialogue with each other, while the AI intervenes only where the conversation requires mediation or context.
01
Build an opinion profile
02
Match across disagreement
03
Speak through the mediation layer
04
Verify and reflect
The mediation architecture can be configured for different forms of structured dialogue.
Matching, mediation, translation, verification and reflection are implemented as separate system functions. By adapting conversation rules, source context and interaction setup, Parallaxe can support different dialogue formats without changing the underlying mediation logic.
When AI mediates a conversation, it also shapes the conditions under which that conversation takes place.
Parallaxe investigates how mediation changes political dialogue: what happens when emotionally charged language is reformulated, vague arguments trigger follow-up questions, claims are checked against external sources and insufficient evidence leads the system to withhold judgment. These interventions can reduce immediate friction, but they also introduce new forms of influence through wording, timing, source selection and evidence thresholds. The project therefore treats neutrality as a system-design problem: when should AI intervene, what should remain untouched, and how can those interventions stay visible and contestable to the people in the conversation?
Public testing informed the next iteration of the system.
Through the Designer in Lab collaboration with Fraunhofer HHI, Parallaxe moved from an initial prototype into scientific evaluation and public deployment. It was presented at Berlin Science Week 2025 and later rebuilt for a larger live-testing format at Futurium, where participant feedback and discussion informed a subsequent iteration of the project.
Award
Recognition
(2024-26©)
