Aporia.
An observatory for structured AI-supported reasoning.
Developed in collaboration with Professor Viktoria Dalko
A reasoning observatory, not an answer engine.
Aporia is a conversational research prototype designed to improve the visibility and discipline of human reasoning. Instead of producing an immediate answer, it organizes inquiry through four observation lenses — Dialectic, Socratic, Reflection, and Transfer — while preserving assumptions, tensions, uncertainty, and the next cognitive move.
Its reasoning indicators are exploratory research signals, not validated educational or psychological assessments.
Most conversational AI rewards speed, completion, and agreement.
Aporia investigates a different interaction model: whether AI can help people inspect how a judgment is formed, challenge assumptions without becoming adversarial, and preserve uncertainty until the evidence supports resolution.
Four observation lenses, one continuous inquiry.
Each lens is a mode of attention. Together they let a single conversation move between building competing interpretations, testing assumptions, watching reasoning change, and carrying that reasoning into a new context.
Build the strongest competing interpretations before synthesis.
Surface assumptions and identify what evidence would change the conclusion.
Notice how and why the user's thinking changes.
Apply the reasoning pattern to a novel context.
From a conceptual engine to a public research prototype.
Aporia has moved through three distinct iterations. Each one exposed a different limitation and pushed the next design toward more structure, more transparency, and clearer research boundaries.
- Prototype 1Conceptual engine
- —Established the core idea of an AI reasoning partner rather than an answer generator.
- —Used a direct conversational flow and early inquiry prompts.
- —Revealed the need for clearer cognitive modes, visible uncertainty, and stronger failure handling.
- Prototype 2Structured observatory
- —Introduced the Aporia identity and Observatory interface.
- —Separated inquiry into Dialectic, Socratic, Reflection, and Transfer lenses.
- —Added live reasoning signals, active tension, next cognitive move, and exportable observatory records.
- Current iterationPublic research prototype
- —Added a deterministic demonstration mode and optional live-model connection.
- —Added input validation, JSON/error containment, and graceful fallback behavior.
- —Added security and privacy disclosures, a public technical report, public repository, and public deployment.
- —Current limitation: cognitive indicators remain exploratory and require empirical validation.
| Dimension | Prototype 1 | Current iteration |
|---|---|---|
| Interaction model | Answer-generating conversation | Reasoning observatory with four lenses |
| Inquiry structure | Free-form prompts | Dialectic · Socratic · Reflection · Transfer |
| Reasoning visibility | Implicit inside the answer | Live signals, active tension, next cognitive move |
| Reliability | Ad-hoc failure handling | Input validation, JSON/error containment, graceful fallback |
| Research transparency | Internal notes only | Public technical report and disclosures |
| Public access | Not deployed | Public deployment, public repository, deterministic demo mode |
A short walkthrough of the Observatory interface.
A 20-second walkthrough of the Observatory interface, dialectical inquiry, live reasoning signals, and the Reflection lens.