Insights From the Winners of the ODSC West 2024 NVIDIA Hackathons
We’ve had hackathons in the past, but the NVIDIA Hackathon during ODSC West 2024 was easily our biggest one yet. In person, 230 teams competed to develop machine learning applications or solutions using NVIDIA’s RAPIDS accelerated Python libraries. Meanwhile, there were 270 virtual teams that competed to enhance the accuracy of a large language model (LLM) for a specific task by processing high-quality data with NeMo Curator.
The winners were:
In-Person Hackathon:
1st Place: Shyamal Shah (L4)
2nd Place: DeepMind Ace: Feifan Liu, Himalaya Dua, Sara Zare
3rd Place: Lorenzo Mondragon
Virtual Hackathon:
1st Place: KeepLearning
2nd Place: Deep Visionaries
3rd Place: Anya Stetsenko
What do the winners have to say?
Hackathons can be intimidating, especially if you’ve never done one before. So, what do some of the winners have to say about their experience with this hackathon, and what advice do they have to share with other prospective competitors?
“Participating in the NVIDIA AI Hackathon at ODSC West was a fantastic experience, full of energy and learning. Our winning approach focused on efficiency — using LoRA for parameter-efficient tuning to maximize performance without overloading resources. We carefully set key hyperparameters and optimized them for memory and speed, creating a stable model that met our performance goals. My advice for future hackathon participants is to stay adaptable and keep the big picture in mind. Hackathons move fast, so being ready to pivot and make quick decisions is essential. Above all, enjoy the process — it’s a unique learning experience, win or lose!”
– Sri Poojitha Mandali
“This was the first time for me working with LLMs, my background is Neuroscience but I was always very curious about AI and wanted to learn about it. It was extremely fun and intense and rewarding, I was digging deep into the techniques for fine-tuning, learned the theory behind it to understand conceptually what is going on, and talked on Discord and Slack with others.”
“At first, I was just improving the data curation and prompt, and also micro-batch size and learning steps (number and interval). But the biggest improvement in fine-tuning came from adding synthetic data, with a caveat that it was important to have the highest possible quality of it, so I was very careful with the data curation for SDG, because as we know: ‘garbage in — garbage out’”
“My biggest advice would be to use Discord and not be afraid to ask questions. Also, even though there are time constraints, I recommend taking time to understand the problem on a deep level, it can give a better perspective for a solution.”
– Anya Stetsenko
“The NVIDIA/ODSC hackathon challenged me to analyze an extensive Halloween-themed dataset using NVIDIA’s powerful GPUs through Google Colab. My approach focused on understanding the data’s underlying structure through careful feature analysis. By identifying multicollinearity patterns and implementing target mean encoding for categorical variables, I was able to narrow down the original 106 features to just three key predictors.”
“My advice is simple: just do it. Don’t let perfectionism or imposter syndrome hold you back. While winning is gratifying, the real value lies in the connections you forge and the knowledge you gain along the way. Every hackathon is an opportunity to grow, regardless of the outcome.”
– Shyamal Shah
What’s next?
While we’re still in the early planning stage of ODSC East 2025, we’re planning big things for the event’s 10th anniversary! This includes a hackathon, among other things, to test your AI expertise. If you’re interested in attending the conference, then you should register now while tickets are still heavily discounted.
You can also sign up for our free weekly newsletter to get updates for any future hackathons!
