Deep Research

Database: alh_deep_research · Repo: alhazen-skill-deep-research

The Deep Research application turns open-ended investigation into structured, queryable knowledge. Its skills discover sources, ingest them, and analyze them — building literature corpora, technology landscapes, and hypothesis timelines in the alh_deep_research database. These are the research engines the Personal Assistant’s analyst commissions when a decision needs evidence.

Skills

Skill	What it does
Scientific Literature	Multi-source literature search and ingestion (Europe PMC, PubMed, OpenAlex, bioRxiv/medRxiv) with semantic search and thematic clustering. Builds the `scilit-paper` corpus.
Tech Recon	Goal-driven technology investigation — interview, discover candidates, ingest sources with parallel subagents, and render Observable Plot analysis against user-defined success criteria.
Literature Trends	Abductive argumentation-based trend analysis: how explanatory hypotheses evolve, are refined, or are abandoned across a tagged literature corpus over time.
DisMech Notebook	A GLAV bridge mapping the standalone DisMech rare-disease knowledge base into the notebook memory model for cross-skill referencing.

The database

All four skills load their namespaces (scilit-, trec-, and the DisMech notebook’s dm- mappings) into alh_deep_research. Literature Trends operates on scilit-paper entities already ingested by Scientific Literature, so the two compose within the same database. Select alh_deep_research in the dashboard switcher to see the Scientific Literature, Tech Recon, and DisMech Notebook dashboards.

Commissioned by the analyst

Deep Research owns discovery and ingestion; it does not decide what a piece of research means for a decision. That is the Research Analyst’s job. The analyst (in alh_personal) frames a decision-linked question, dispatches it to Tech Recon or Scientific Literature here, and reads the resulting investigation back — holding the cross-database link as an external_ref (e.g. a trec-investigation id) rather than a graph edge. See the Multi-Database Model for how that soft reference works, and the Personal Assistant for the commissioning side.