research-infrastructure

the overarching system that holds:

• learning-dashboard
• project-supervisor-bot
• study-companion-bot
• confidence-bot

as well as connecting to my obsidian vault and the website (the phd-live platform)

progess and updates kept here: system-map

the obsidian vault acts a basis that feeds into an intrinsic shared-context-layer which acts as additional contextual information for all components of the research infrastructure.

following session 2026-04-17, the ‘research infrastructure’ was born.

Research / knowledge infrastructure: key decisions

(Off-the-shelf options (LibreChat, AnythingLLM, Chainlit) don’t accommodate the non-chat interaction paradigms that I am interested in exploring).

general logic - don’t build ‘plumbing’ that already exists; but build the interface layer myself

The three bots (planned so far)

• Supervisor bot — structured critique, accountability, session-based. v1.5 working.
• Study companion — exploratory thinking, document-aware, peer/interlocutor role.
• Confidence bot — public presence agent. Reads notes, proposes shares, approval
  step retained. Most conceptually distinctive -> next to build?

These map onto three modes of cognitive/emotional labour in PhD research. -> This is worth naming explicitly in the thesis framing.

shared context layer

One vector store (ChromaDB or SQLite + embeddings), one query API, all bots consume it. Sequence: watcher → embed → query endpoint → wire supervisor bot → build confidence bot on top. Chunk by Obsidian heading structure.

(RAG-compatible by design; fuller retrieval can be added later as extension).

learning dashboard

Shell first (navigation + activity feed), which can become knowledge layer second.

I will let real usage generate signal before designing potential context features of this dashboard.

Document this sequencing decision - the reasoning is important research material (ie. why starting from shell first, seeing what emerges as I use it before adding more features)

Activity log (exists only on phd-live site? or also on the learning dashboard)

Three distinct things, kept separate:

• Session record — structured log per bot interaction (local store)
• Research diary layer — interpretive, themes, trajectory (future)
• Public feed — curated subset surfaced to PhD-Live (separate concern)

Open questions

• Does the learning dashboard become a context source for bots, and when?
• What is the “shareable” threshold logic for Confidence Bot?
• Does the activity log feed PhD-Live automatically or via curation?