Connectome treats curiosity and confusion as first-class signals. Instead of static Q&A, users annotate what felt unclear or interesting, rank intensity, and watch those signals shape the next response. Structured extraction pulls feedback into a persistent topic graph that accumulates across sessions—a navigable map of your own learning trajectory.
Built as an API-first, human-in-the-loop LLM orchestration layer. Core patterns: feedback-driven personalization, session persistence, EWA-based relevance decay, and force-directed visualization. Designed to explore how memory, topology, and feedback can support open-ended learning without badges, streaks, or dark patterns.
Stack: FastAPI, Claude API, SQLite, D3.
You wander. It maps. Side effects may include genuine curiosity.
These patterns generalize beyond learning:
| Pattern | Implementation | Generalizes to... |
|---|---|---|
| Structured extraction | Parse <topology> JSON from natural language responses |
Document intelligence, entity extraction |
| Feedback injection | Annotations modify next prompt context | User corrections, review queues, preference learning |
| Session persistence | Topic graph accumulates in SQLite across conversations | Customer history, case continuity, audit trails |
| Relevance decay | EWA-weighted opacity based on visit recency | Lead scoring, stale flag detection, attention prioritization |
| Human-in-the-loop | Signals steer without blocking | Escalation triggers, QC sampling, compliance review |
- Dialogic conversation — an AI that thinks alongside you, not at you
- Annotation feedback — highlight text to mark "confused" or "curious," shape the next response
- Topology visualization — force-directed graph of your learning trajectory
- Honest engagement — no "Great question!", no dark patterns, no gamification
Your exploration builds a persistent graph:
| Visual | Meaning |
|---|---|
| Circle radius | Time/depth spent on that topic |
| Opacity | Recency (EWA of visit intervals) |
| Pink fill | Net confusion from annotations |
| Green fill | Net curiosity from annotations |
| Dashed outline | Adjacent topic (suggested, not yet explored) |
This is the honest reward: not fake coins, but the actual shape of your intellectual growth.