How We Organized an AI Hackathon: 15 Lessons Learned
It was back in 2004 when I had my very first hackathon. 24 hours of Coding. Creativity. Fun.
Non-engineers don’t usually get the idea behind hackathons. A lot of them think that it’s a real torture to sit down and do something for 24 hours with no sleep or rest. In fact, hackathons are super fun for people who love to challenge themselves, contribute to interesting ideas and of course it is a great place to make friends.
There is a strong hackathon culture at PicsArt. Usually, we are hosting 1–2 hackathons per year within the company and encouraging our folks to work on something that is not directly related to what they are working on right now. And this is how we got our first AI solution back in 2015!
It was related to the server side style transfer. It was very slow, but it had amazingly good results. So we didn’t just stop there.
From 2016, we started actively investing in AI. We are now organizing regular training events, and we don’t spare money on AI research. We have launched an AI lab in Moscow.
And guess what? There came a point when we decided to do an AI hackathon!
Well, that’s easier said than done. Cause organizing such an event is a real pain in the ass!
AI hackathons differ hugely from usual hackathons in a few ways. For example, AI solutions take more time for research, gathering the data, training models, retraining, etc. It is hard to imagine that one can get something tangible within 24 hours of hacking.
And it happens that there are no best practices for AI hackathons yet.
We have visited several AI hackathons and we had a chance to attend a pretty good one in Minsk once.
So, there came a moment when we realized that we had done enough research and we had enough stamina to try and organize our own AI hackathon. We picked Moscow as an event destination, and we had a few reasons for that. First of all, we wanted to discover new talent, and Russia is a great place for that. Also, we have an AI lab there, and this event would be great for networking and getting to know the IT landscape there.
So, as we started working on the plan, we felt that we had to adjust the standard hackathon format to match it with our needs by using our learnings from the research.
We decided to start on Friday 4 PM and finish by Sunday 6 PM with all the final presentations to utilize max free time of the week, we set a strict 60 seconds pitching time for teams at the beginning and planned to provide both business and tech mentors.
Also, since we were new in Russia, we decided to offer great prizes to really excite the participants.
Since we were anticipating that a significant number of applicants would flow in, we decided to start with an online competition which offered small prizes and was meant to help us select the applicants for the actual AI hackathon.
It was also possible to apply if you already had a team. The teams had only to show a prototype or a polished idea at least. So there were basically 3 ways to enter the hackathon:
- The tech guys could participate by competing online
- Product teams had to present their ideas
- Established teams had to show prototypes
We found great event organizers who helped pick a cool location, decorate and conduct our event. The guests had to arrive at a walking district full of bars and clubs, and we had people guiding the guests to the event venue.
We had great teams of hackers ready to rock their projects: some were really challenging to do! And the guys did it.
Our collective learnings
Anything you do can be perfected. Our AI hackathon brought great results, but if we are to organize another one, we already know what to focus on, what to change, or what to leave out from the plan.
Below are the collective learnings we got from our team members, participants, and organizers.
Make sure the results meet the jury’s expectations
- Be very clear with the direction and the scope of your hackathon to make sure the results meet the jury’s expectations.
- There will always be teams that will try to use ready GitHub solutions. So, rules should be clear there.
- There will be teams with legacy projects to which they want to add an AI solution; It would be much fairer if you separate such teams from the freshly-organized ones and nominate them for a separate prize category.
Pre-selection competition should be exciting and challenging
4. The tasks for the online pre-selection competition should be exciting and challenging at the same time, and they should by no means be available on GitHub or other public places.
5. Evaluation criteria should be based on a variety of things such as the quality of the results, the architecture of the solution, performance, etc. But keep in mind that this can be time- and resource-consuming.
You might need a few months to get things properly organized
6. We had very little time to prepare for this one. But overall, you might need a few months to get things organized: starting from the online competition and preselection and ending with the actual hackathon.
7. Should you have an online part in the hackathon, consider sending out a preliminary notification about the winners. Also, consider the gap between the end of the online phase and start of the main event to avoid logistic complications as there may be participants from different regions
8. We started on Friday at 4 pm which was not a good idea since it takes a bunch of time for people to get to a destination in Moscow on a Friday afternoon. We will shift at least a couple hours next time.
Internet connection is CRITICAL for AI projects
9. Internet connection is CRITICAL for AI projects so make sure to have the best and the fastest WiFi ever! And have a plan B in case something goes off your schedule.
10. It would be great to have GPU resources for all, they are costly of course.
11. Ready datasets which participants can use for their projects will improve result quality drastically.
12. Instead of developing our own website, we could have used the existing hackathon builders which can be easily found in the market. This would have saved a lot of time!
You are going to meet great talent
13. You are going to meet great talent. Having a talent acquisition plan in place would be excellent.
14. The tech mentors should not leave during the night. If the hackers are there, the mentors should be there too.
15. Mentors should be in sync when giving advice and be very careful not to over- influence the final results.
Final notes
So, what did we get as a result? As a company, we got to know bright people, great AI enthusiasts, and innovators. And as hackathon organizers, we leveled up by rocking yet another hacking event.
Let’s see what comes next!
Stay tuned for more PicsArt Stories.
About the author
Artavazd Mehrabyan is a tech advocate and entrepreneur with an engineering background and extensive experience in the startup world. Currently he is the Co-Founder and Chief Technology Officer at PicsArt, the leading creative platform for social media storytelling.
