DOIM Explorer

An unofficial, public-data explorer for the University of Utah Department of Internal Medicine. It presents the department's official division directories, public faculty profiles, and (after a separate collection step) scholarly publications.

Refresh the public faculty directory

uv sync --locked --all-extras --dev
uv run --locked doim-directory --collect --strict

The command writes matching versioned documents to data/directory.json and site/data/directory.json. It honours robots policies through research-explorer-core, records source health for all 12 divisions, and refuses an unsafe total or per-division count drop.

Faculty profiles are collected separately from publications. Review the directory diff before running doim-publications; publication collection uses the generated, University-of-Utah-affiliation PubMed queries in the directory document.

Incremental faculty refresh

To refresh one or a few accepted faculty records without re-querying the whole department, provide their stable U of U IDs. The command refetches only those public profiles, applies their overrides, collects all supported publication sources for them, and rebuilds the complete retrieval index while reusing vectors for unchanged chunks.

export RESEARCH_EXPLORER_CONTACT_EMAIL='[email protected]'
GOOGLE_CLOUD_PROJECT="$GCP_PROJECT" uv run --locked doim-refresh --faculty-ids u6039184,u0012345

The command requires accepted data/directory.json, publication documents, and data/rag/ as its base. It stages every generated artifact before publishing, retains prior active-faculty publications until the normal retention limit, and refuses unknown IDs, a profile/source error, or a missing PubMed contact address. Review and commit its directory, publication, static-site, and RAG changes together.

Collect publications

PubMed's E-utilities identify every caller by a contact address, so export one before collecting. Without it each PubMed source is recorded as skipped and the run reports only the publications already retained in the snapshot.

export RESEARCH_EXPLORER_CONTACT_EMAIL='[email protected]'
uv run --locked doim-publications --strict

The command writes data/publications.json, data/publication-details.json, and their site/data copies, merging newly collected works into the retained snapshot. Because the published retrieval index records the directory and publication timestamps it was built from, rebuild and commit it whenever either document is regenerated:

GOOGLE_CLOUD_PROJECT="$GCP_PROJECT" uv run --locked doim-rag

Build the collaboration network

The Network view draws the co-authorship graph between DOIM faculty: an edge exists whenever the publication pipeline attributes the same work to two of them. Regenerate it whenever the directory or publications documents change:

uv run --locked doim-collaboration

The command reads the accepted data/directory.json and data/publications.json, and writes matching versioned documents to data/collaboration.json and site/data/collaboration.json. Only faculty with at least one internal collaborator become nodes, and only where their connected group holds at least five people — smaller pockets crowd the canvas without showing a structure worth reading. Node color groups a division using a palette validated for both color-vision-deficient and full-color readers, and dot size reads collaborator count on a cube root (see doim_explorer/collaboration.py and site/assets/app.js). Layout is precomputed and deterministic, so regenerating from unchanged inputs never moves a dot. doim-refresh (above) publishes this document automatically as part of its own staged commit, so this command is normally only needed after running doim-directory/doim-publications directly. See docs/doim-data-contracts.md for the document's fields and limitations, and site/assets/vendor/VENDOR.md for the one third-party library the site loads (vendored locally, not from a CDN) to draw it.

Fellows, residents, and Networks +

The Networks + view adds the program's fellows and residents, listed in config/trainees.toml. Each is searched in PubMed like faculty, on their full author name at a University of Utah affiliation, and the network is rebuilt from both sets of publications:

uv run --locked doim-trainee-publications --strict
uv run --locked doim-network-plus

The weekly directory refresh runs both. When the program sends a new roster spreadsheet, regenerate the config (e-mail addresses are dropped; hand-added also_known_as names and pubmed_query overrides of people still on the roster are kept). The next search drops the publications of people no longer on it:

uv run --locked --extra trainees doim-trainee-roster "<roster>.xlsx"

Configuration

config/directory.toml is the active source manifest. It lists the official department/division URLs, affiliation terms used for PubMed queries, and the publication collection policy. config/faculty-overrides.toml holds sparse, reviewed corrections and enrichment keyed by University faculty ID; it never duplicates the roster. See docs/doim-faculty-overrides.md for the editing and refresh workflow. config/branding.toml contains the public site identity.

The reusable collection, publication, and retrieval implementation is provided by research-explorer-core v0.2.1. The local sibling checkout at ../research-explorer-core is the same release commit pinned by this project; keeping the Git pin makes installs and GitHub Actions reproducible.

Historical InsightNet configuration and workflows are retained only under legacy/insightnet/; they are not part of the active DOIM refresh.

GitHub Actions

ci.yml runs the locked environment, Ruff, and the full test suite on pull requests and pushes to main. refresh-doim-directory.yml reconciles division rosters weekly and refreshes every public profile on the first day of January, March, May, July, September, and November. Each run re-queries publication sources for every faculty member, so new papers arrive weekly; a source error for someone whose publication inputs did not change keeps their previous works, and the run fails only for errors on changed faculty or an outage (more than 10% of the rest). It rebuilds the RAG index incrementally (only new chunks are embedded) and opens one reviewable artifact pull request. refresh-doim-faculty.yml is the manual equivalent: enter comma/whitespace-separated faculty IDs after their override is on main. Both workflows require the non-secret RESEARCH_EXPLORER_CONTACT_EMAIL repository variable. Once a generated pull request is merged, its site/data change starts deploy-pages.yml for the reviewed main content.

deploy-doim-ask.yml uses Workload Identity Federation (WIF) to publish and deploy the DOIM Ask image when its application or retrieval-index inputs change. Run verify-doim-auth.yml manually whenever a Google Cloud or GitHub Actions setting changes; it verifies that the WIF-backed deploy identity can inspect the DOIM Artifact Registry repository and Cloud Run service without using a long-lived key.

After Pages and Cloud Run have rolled out the same accepted commit, run verify-doim-release.yml. It checks source health, generated-document schema and publication identity, the committed retrieval-index provenance, the public Pages documents, Cloud Run readiness, and the browser CORS preflight without sending an Ask request to Vertex AI. The equivalent local command is:

uv run --locked doim-release-check