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.

Data

The conference list lives in data/conferences.json: one record per conference per year, inside { "conferences": [ ... ] }. The file is bundled into a Cloudflare Worker at deploy time and served from memory. There is no database: the API reads this file, and corrections are pull requests.

Data model

A conference record is one meeting of a series in one year, such as ISMB 2027. Its deadlines are the dated milestones before the meeting, such as the abstract or paper deadline.

Field	Type	Required	Meaning
name	string	yes	Short series name, the same every year, e.g. ISMB
year	integer	yes	Year of the meeting
full_name	string		Longer name, e.g. 18th ACM BCB
location	string		City and region or country
start_date, end_date	date		First and last day, YYYY-MM-DD
link	URL		Official page for this year's meeting
deadlines	list		Milestones, each { "name", "date" }. name is a label such as Abstracts or Full papers; date is YYYY-MM-DD
note	string		Free-text note
attending	boolean		Lab-internal: the lab plans to attend
slug	string		Overrides the series ID the API derives from name

Example:

{
  "name": "ACM BCB",
  "full_name": "18th ACM BCB",
  "year": 2027,
  "location": "Boston, MA",
  "start_date": "2027-08-10",
  "end_date": "2027-08-13",
  "link": "https://acmbcb.hosting.acm.org/",
  "deadlines": [
    { "name": "Abstract preregistration", "date": "2027-03-19" },
    { "name": "Full papers", "date": "2027-03-26" }
  ]
}

From each record the API derives a series slug (acm-bcb), an instance id (acm-bcb-2027), and a deadline id (acm-bcb-2027:full-papers). The /deadlines feed also lists each meeting's start date as a Conference deadline. The JSON Schema is schema.json. The updater's research notes are kept separately in automation/research-state/ and are not served.

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 GitHub Actions workflow, scheduled-conference-update.yml, runs Claude Code every Monday at 09:00 UTC to refresh the data. It can also be started by hand from the Actions tab. 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 works on the shared conference-update branch and opens or updates one "Conference update" pull request for human review; nothing merges on its own. Claude does the research; when it finishes, a hook runs scripts/publish-update.sh, which runs the checks (failures go back to Claude), commits, pushes, and opens the pull request as claude[bot]. Merge or close that pull request before the next run, because every run updates the research-state files.

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).