I was thinking about the second -athon plans of terminology-mapping-athon) as follow up on Mind meet machines first *-athon. Mainly if it is ambitious enough given current state of tools like Codex, Claude code/cowork, Antigravity, etc…
How about vibe-athon. Ideally mixed teams (human(s)+agents) but we can have some for agents only. For agents only, those are harder to define fully.
And if not done as event during an OHDSI symposium, it can be done as “ongoing” event or as topic-designated month (vibe coding March or May or June)
This is the proposal:
Community formulates ideas and puts them in a pool of ideas for vibe-coding later by others.
Some are selected for the event (others can be left as inspiration for later; “parking lot”)
There is a competition or group activity around best solution for an idea.
We focus on smaller ideas rather than projects that require 20 days of agent coding (like the recent record of a new C compiler)
A group of humans evaluate/judge the success of each team on an idea. (or for some ideas we may have automated way of judging)
It may be too far fetched (too ambitious; not quite there yet with tools) but the trick is to come up with the right size of the challenge (not too ambitious)
As cherry on top, winning team(s) are invited by a company to possibly put some business spin on it. Well, others may prefer keeping it purely open source.
Possible ideas:
(to seed some ideas, please contribute yours below) (similarly to recent activity around StudyAgent)
- Rewrite DQD package in Python (use current R code and documentation as input)
- Make improvements in CapR package (look at all issues in the repo currently)
- Refactor Circe and other conundrums in Atlas (I think there is some Java dependency)
- Write fancy tool that guides a site in joining EvidenceNetwork
- Suggest JSON schema for Cohort (like TAB did recently for ConceptSet) and evaluate it on published cohorts, Phenotype library cohorts, Atlas-defined cohorts)
- write CapR version of all Phenotype library cohorts and/or of other cohorts (in past OHDSI studies in repo or published in pubmed indexed journal)
- rewrite some OHDSI tools such that they run in AllOfUs workbench (e.g., Achilles characterization considering not having DatabaseConnector as an option (connectionDetails is not an option)