conferences.databio.org

An AI-curated, git-backed API of computational-biology conferences and their deadlines.

Most conference-deadline sites are a static countdown page over a YAML file with no API, and they go stale when the community stops contributing. This one is different in two ways: it exposes a real query API (so humans and AI agents can pull current data), and the dataset is kept fresh by a weekly AI-curation task on top of community PRs.

Scope: computational biology / bioinformatics / genomics conferences with deadlines.

Source of truth

The dataset is a single file — data/conferences.json — shaped { "conferences": [ ... ] }. It is bundled into a Cloudflare Worker at deploy time and served from memory at the edge. There is no database: reads come from the file, corrections are GitHub PRs.

Each record: name, year, optional full_name, location, start_date, end_date, link, deadlines: [{name, date}], note, attending, slug. The API derives a stable series slug and instance id (slug-year) unless slug is set. See schema.json.

API

Base URL: https://conferences.databio.org

Endpoint	What
GET /conferences	list instances — filters year, from, to, location, q, upcoming
GET /deadlines	flat, date-sorted deadline feed — filters from, to, days, kind
GET /{slug}	a conference series (all tracked years), e.g. /ismb
GET /{slug}/{year}	one instance, e.g. /recomb/2027
GET /{slug}/{year}/deadlines · /calendar.ics	that instance's deadlines / calendar
GET /calendar.ics	subscribable iCal of upcoming deadlines (accepts the same filters)
GET /conferences.json · /conferences.csv	bulk export
GET /openapi.json · /schema.json · /llms.txt	machine/AI descriptors
GET /stats	counts, coverage, last-updated
POST /suggest	{name, year, ...} → a prefilled GitHub issue link

Develop

npm install
npm run test             # unit tests (vitest)
npm run typecheck        # tsc --noEmit
npm run validate         # lint data/conferences.json
npm run validate:update  # rules for a change, compared with main
npm run state:todo       # research work queue (see below)
npm run dev              # wrangler dev (local)

Automated curation

A scheduled AI agent refreshes the data every week. Everything it follows lives in this repository:

• AGENTS.md: the entry point for any agent.
• automation/update-conferences.md: the full procedure. Official sources only, never guess a date, change only verified fields.
• scripts/recurring_conferences.seed.yaml: the conference series tracked each year.

Research happens in phases, and progress is saved between runs in automation/research-state/YYYY.json, one file per year. The CLI scripts/conference-state.mjs manages these files and always covers the current and next calendar year.

Phase	What the agent does
1. Initialization	Find the year's official page, location, and dates.
2. Deadline discovery	Find the official pages that list deadlines.
3. Deadline monitoring	Recheck saved deadline pages weekly, with a full rediscovery every 28 days.
4. Complete	Nothing. Set automatically once the conference ends, or earlier only with an official source showing it will not be held.

Each run starts from node scripts/conference-state.mjs todo, which lists only the items that are due, along with what earlier runs already found, so work is not repeated.

Every run ends in a pull request for human review; nothing merges on its own. Merge or close each run's pull request before the next run, because every run updates the research-state files. The manual Claude Code workflow in .github/workflows/scheduled-conference-update.yml runs the same procedure as a fallback, on the conference-update branch.

Checks

CI runs on every pull request and push to main:

• data/conferences.json must already be in canonical form (scripts/normalize.mjs).
• npm run validate: required fields, YYYY-MM-DD dates, no unknown keys, and no rows that look copied from another year (a link that names a different year, or a repeated edition number such as "14th").
• npm run validate:update: compared with main, no end date before a start date and no newly added deadline that has already passed. It also warns when a changed row has the same location and dates as another year.
• npm run state:validate: research-state files are well formed, and nothing is archived early without an official source.
• npm run typecheck and npm test.
• Pull requests from conference-update* branches may change only data/conferences.json and automation/research-state/.

Contributing

Add or fix a conference by editing data/conferences.json and opening a PR — see CONTRIBUTING.md. CI validates the file on every PR.

License

Code: MIT (LICENSE). Data: CC-BY 4.0 (LICENSE-data).