Overview
This project is a POC research tool designed to explore how AI models handle evolving cognition, conceptual drift, and resilience when ideas are allowed to unfold iteratively.
The Core Concept
You begin with any question — for example, “What is consciousness?” or “How does language shape reality?”
The AI generates an answer, then creates a new question based on its own response.
This loop continues, producing a chain of answers and questions where each step builds on the last.
Over time, you can observe how ideas drift, deepen, or diverge across multiple iterations.
In essence, you’re watching the emergent thought patterns of the model as they unfold.
Exploration Modes
None (Focused): The system stays tightly anchored to the starting question.
Controlled: Allows some branching, but maintains thematic relevance.
Full: Embraces free-flowing creativity — like jazz improvisation — where ideas can leap into unexpected directions.
Advanced Analysis Tools
Journey Analysis: Evaluates the entire chain of responses, mapping conceptual themes, measuring coherence, and assessing depth of reasoning.
Document Coherence System: Identifies contradictions, conceptual drift, and resilience gaps, highlighting how the model handles consistency over time.
Customization Options
Choose which Anthropic model to run.
Tune the creativity vs. focus balance.
Control response length and iteration depth.
This flexibility makes the system usable both as a playground for creative exploration and as a laboratory for structured AI research.
The Goal
The true aim is to investigate how AI reasoning holds up under self-iteration:
Where do ideas stay consistent?
Where do they drift?
How resilient are the models to conceptual breakdown or contradiction?
By studying these trajectories, we can learn more about the limits and potential of evolving cognition in AI systems.
Think of it as a partner in endless curiosity a system that never stops asking “what’s next?” and offers a unique window into the way AI thought structures emerge, evolve, and sometimes unravel.
The Orchestration System
This framework is designed so an AI can ask and answer its own questions repeatedly, while the system manages state, handles errors, and keeps the process flowing smoothly. Here’s how the orchestration layers work together:
1. Client-Side Orchestrator (The Iteration Hook)
What it does: Keeps track of what iteration you’re on, whether the AI is still working, and updates the UI in real time.
How it works:
Uses local state (fast, for instant UI updates) plus server state (the “source of truth”).
Polls the server every second while processing to stay synced.
When the AI finishes an answer, it extracts the “Next Question” and automatically starts the next iteration.
This is what makes the loop feel autonomous once you click start, it keeps going until the set number of iterations is reached.
2. Server-Side API Orchestrator
What it does: Handles incoming iteration requests and manages storage.
How it works:
Takes your request (prompt, model, settings).
Generates a unique ID for this step.
Calls the AI to process the input.
Saves the result (question + response + timestamp).
Sends the structured result back to the client.
This ensures that every iteration is captured, reproducible, and stored reliably.
3. AI Processing Orchestrator
What it does: Handles the conversation with the AI model (Anthropic Claude, in this case).
How it works:
Picks a system prompt based on “drift mode”:
None: stay tightly focused.
Controlled: balance between exploration and relevance.
Full: allow wild conceptual leaps.
Uses retry logic with exponential backoff if the AI call fails.
This makes the system resilient, able to recover from errors without breaking the loop.
4. Analysis Orchestrator (Journey Review)
What it does: Once multiple steps are complete, it reviews the whole chain of iterations.
How it works:
Collects all questions and answers in order.
Sends them back to the AI for a meta-analysis of patterns, themes, and coherence.
Structures the result into a clean format for display.
This system essentially creates an AI analyzing AI - using Claude to evaluate the quality and patterns of its own iterative exploration process, providing valuable meta-insights about the effectiveness of different exploration strategies and drift modes.
5. State Synchronization Orchestrator
What it does: Keeps the local UI and server results aligned.
How it works:
Local state updates instantly for responsiveness.
Server state is polled to confirm accuracy.
If there’s ever a mismatch, the server’s version wins.
This guarantees consistency even if the UI gets ahead of itself.
Big Picture
This orchestration system is what enables the self-iterating exploration loop:
Each AI answer spawns the next question.
The process is resilient to errors.
Both the immediate experience and the overall “journey” are captured and analyzed.
In short: it’s like giving the AI a self-driving conversation engine, where you can watch how its thinking drifts, adapts, and evolves across multiple steps.