SOLYSTOPIAProject Hearthmind
Hearthdata Foundry
Turning household-specific documents and records into material the local models can actually use.
Progress
0% complete
Objective
What this quest sets out to do
Build a simple, private pipeline for feeding household-specific documents (manuals, records, notes) into the local system as retrieval context, without sending that data to any external service.
Why it matters
Why Solystopia cares
Generic model capability is not the same as useful household capability. This quest is the difference between "a chatbot" and "a system that knows this household's actual documents" — while keeping that material inside Hearthvault.
Components
What this is actually made of
Milestones
Where this quest stands
2026 Q2
Local embedding model selected
Planned2026 Q2
First household document set ingested
Planned2026 Q2
Dataset inventory with provenance and licensing metadata completed
Per the evidence base's section 9.2 research priorities.
Planned2026 Q3
Evaluation design defined for measuring fine-tuning gains
Per the evidence base's section 9.2 research priorities.
PlannedOpen design questions
What is not yet decided
- Is a full vector database overkill at household scale, or does simplicity actually argue for one anyway rather than a simpler search index?
- What evaluation methodology would actually prove that fine-tuning on household-specific data improves real task performance, rather than just plausibly should?
- How should the household draw the line between what it fine-tunes on versus what it only retrieves via RAG?