{
  "$type": "site.standard.document",
  "bskyPostRef": {
    "cid": "bafyreigiyhab2lm3xravpup3hjcvfazp5rlw3h5zafkqxrhkjazkc7q4be",
    "uri": "at://did:plc:pgryn3ephfd2xgft23qokfzt/app.bsky.feed.post/3mhbifnfsjc32"
  },
  "path": "/t/splitmind-ai-modeling-llm-replies-as-competing-internal-pressures/174358#post_1",
  "publishedAt": "2026-03-17T15:17:59.000Z",
  "site": "https://discuss.huggingface.co",
  "tags": [
    "github.com",
    "GitHub - yatarousan0227/SplitMind-AI: Psychodynamic-inspired AI agent architecture that..."
  ],
  "textContent": "Hi everyone,\n\nI wanted to share SplitMind-AI, an open-source project exploring a different way to structure conversational LLM systems.\n\nInstead of treating persona as a single prompt layer, SplitMind-AI models reply generation as a negotiation between competing internal pressures: desire, inhibition, defense, norms, and persona integration. The final response is generated from that tension rather than from tone alone.\n\nThe motivation is not psychological realism in a strict sense. It is inspectability. When a response feels off, I want to know whether the issue came from internal pressure, containment, safety constraints, or persona framing.\n\nThe current project includes:\n\n  * a Streamlit interface for chatting and inspecting traces\n  * explicit state for relationship, mood, drive, inhibition, and memory\n  * persistent vault-backed memory\n  * typed contracts between runtime nodes\n  * safety checks, output linting, and scenario-based evaluation scaffolding\n\n\n\nIt is still a research/architecture project rather than a polished end-user product, but I’d love feedback from people working on agent design, evals, and controllable generation.\n\nRepo:\n\ngithub.com\n\n### GitHub - yatarousan0227/SplitMind-AI: Psychodynamic-inspired AI agent architecture that...\n\nPsychodynamic-inspired AI agent architecture that generates responses from structured internal tension instead of a single persona prompt.\n\nQuestions I’m especially interested in:\n\n  * Is this kind of decomposition actually helpful for controllability/debuggability?\n  * How would you evaluate “relational texture” or indirect emotional expression?\n  * Where would you draw the line between explicit rules and learned behavior?\n\n\n\nThanks for taking a look.",
  "title": "SplitMind-AI: Modeling LLM replies as competing internal pressures"
}