I have helped organize DevOpsDays Austin for the last 4 to 5 years. If you do not know, DevOpsDays is a volunteer, community-run conference where technology professionals come together regionally to learn about the ecosystem they work in daily. (You should join us: Register today!)

The best part of DevOpsDays is the "open spaces," where it is an unconference style for half the day and people learn things together. I have learned more about the innovative technologies we use, and I have learned about a tool that streamlined my workflow.

There are talks throughout the day, and my primary responsibility in organizing DevOpsDays Austin is the "CFP" (call for papers) Lead, where I help recruit and run the agenda for that part of the conference. We get 100s submissions each year, and it takes a significant amount of time to sift through them. I have even grown a "sixth sense" of reading some of these CFPs to find "sales pitches" or "not in the spirit" of what we are trying to do at DevOpsDays.

Over the last couple of years, we have noticed an increase in AI-written submissions, and it is becoming a real problem. Seriously big problem. It is just to easy to ask an LLM "to write me an abstract or CFP about why the DevOps process is the best way to run production servers" and then copy, paste, and sunmit it.

We will not accept these talks as soon as we realize their CFPs are “robotic,” but we do have to read every one, and honestly, it is a waste of everyone's time. We want our speakers to write something about themselves and their experience so you can help grow our community's knowledge.

Along with AI-generated abstracts, another type of CFP we deprioritize is obvious "sales pitches." Our audience does not want to give you our stage for you to attempt to sell your product. Don’t get me wrong, we have sponsorship time slots where our community understands it is the correct place for this place interaction, but not our main stage or our talk tracks. We also get a few attempts from marketing teams and developer relations professionals to sneak in these types of CFPs, which are just product pitches.

This brings us to my release of ai-abstract-classifier. I have created a wrapper application that takes the abstracts submitted to our conferences and gives a score of 0-100% of being "AI-written" or a possible "sales pitch." This score helps the CFP committee to de-prioritize talks we are not looking for.

This ai-abstract-classifier runs locally, leverages Granite via Ollama, with an interface of AnythingLLM. It is a straightforward, secure, local, and private way to leverage an open source LLM to save us hours of reading.

If you want give it a shot, it has support for CSVs and some test abstracts are already in place, but honestly the main power of this classifier is running the abstracts through our Granite model, ultimately saving us so much time.

Learn more about the Granite models in the IBM Granite Community. Learn how you can use the Granite Models locally in these IBM Developer tutorials:

Using Granite to help organize conferences (and more!)

Using Granite to help organize conferences (and more!)