From February 26 to February 28, 2025, Cere teamed up with CEF.ai to host a hackathon at 42 Berlin. We brought together 22 talented students and industry mentors from the CERE / CEF teams to push the boundaries of AI and Web3 technologies through hands-on collaboration. Today, we’re going to spotlight three standout projects from this event: an AI-powered conversation analysis system, a streamlined Developer Console funding solution, and a cutting-edge computer vision toolkit for media analysis. These innovations showcase the creativity of our participants and demonstrate the possibilities for the future that Cef.ai and Cere Network are building.

Track 1: Decoding Communities with NLP AI

As datasets for machine learning grow at an increasing pace, so do the computational costs. To stay agile and make sure resources are spent wisely on high impact accurate insights, it is important to be able to separate the signal from the noise. That’s exactly what the team led by Jakub Gil set out to achieve in the NLP AI track. Their project, detailed in our Conversation Analytics Wiki, built an AI system to analyze messaging data from platforms like WhatsApp and Telegram. In near-real time, they were able to turn a flood of chat messages into clear, actionable insights about community priorities, who the thought leaders are, and which topics drive the most activity.

‍

Using open-source language models like Llama 3.3 and Deepseek, the team tackled the challenge of processing dynamic datasets as conversations unfolded. They achieved impressive results, with clustering scores (ARI) reaching as high as 0.324, meaning their system could effectively group related messages. Helen Sirenko, a participant, shared her excitement: “It was incredible to tweak models and see the insights sharpen—suddenly, we could spot key contributors and hot topics in real-time.” This prototype isn’t just a proof of concept; it’s a preview of how AI can intuit the nuances of group behavior to separate the signal from the noise. This is the first step to reducing resource costs and increasing insight quality. Jakub will share the standout scripts and models from this track, with open-source contributions soon to be available for the community to build on. Updates to the wiki, featuring performance metrics and insights, will follow shortly after, enriching our knowledge base. Feedback from participants like Helen will also be compiled by March 14, 2025 to refine the prototype and explore further development opportunities.

Track 2: Simplifying Developer Funding with a Smarter Top-up

Reducing developer friction is the top priority for an onboarding flywheel. Funding an account on our Developer Console should be a smooth refined process. The team led by Ulad Palinski took on this challenge, crafting a solution to make topping up accounts seamless using USDC and fiat currencies. Detailed in the Developer Console USDC/Fiat Onboarding Wiki, their project integrated test environments of payment providers like Stripe and built a custom smart contract to securely map deposits to user accounts.

‍

The result is a user-friendly process that streamlines and simplifies Web3 funding. Participants tested the flow, ensuring developers could deposit funds with just a few clicks. This is a big leap from the current bridge-based system. “Seeing it work felt like unlocking a door for new developers,” said one team member. This innovation will lower barriers, making it easier for creators to dive into our ecosystem and build with confidence. Ulad will submit the code contributions, including the custom smart contract and payment integration scripts, for review. Updates to the documentation, featuring best practices and integration guides, will be shared by March 28, 2025, to help onboard future developers.

Track 3: Seeing the World Through Computer Vision AI

What if you could instantly tag and search objects in photos or videos with pinpoint accuracy? That’s the vision the team led by Brett Butterfield brought to life in the Computer Vision track. Using open-source AI models, they developed a system for real-time object identification and tagging, as outlined in the Computer Vision Execution Log. From detecting objects in live streams to generating searchable metadata, their toolkit opens up a world of possibilities for media applications.

‍

Participants like Stepan and Nikolai fine-tuned models to handle diverse datasets, overcoming challenges like computational limits to deliver impressive results. “It was thrilling to see the model pick out objects we’d trained it on—it’s like giving machines a new sense,” Stepan noted. This project lays the groundwork for smarter media search and analysis tools, opening the door for revenue-driving industries like retail, advertising, and big data to get involved. Brett will deliver the computer vision models and scripts, making them available for further exploration. Updates to the wiki, including optimization techniques and model insights, will soon follow, enriching our internal knowledge base. Feedback from participants will also be gathered by March 14, 2025, to shape the toolkit’s future and identify development opportunities.

A Launchpad for What’s Next

These three projects—born in just three days—demonstrate the power of collaboration in a rapidly growing field. The NLP AI system could make community insights more powerful and accessible to all sizes of teams. The Developer Console top-up solution will make it easier to welcome more builders to our platform. The Computer Vision toolkit could bring new kinds of interest from revenue-driving industries. Martijn will share an event recap summarizing key achievements, participant stats, and a video montage of demos and mentor sessions, available in the Hackathon Wiki and on our social channels. The dev team, including Jakub, Ulad, and Brett, is already reviewing participant contributions for integration into Cere Network products. Ulad will also update the hackathon wikis by March 28, 2025, capturing best practices and insights from each track.

‍

The hackathon at 42 Berlin wasn’t just an event; it was a spark. A huge thanks to our participants, mentors, and the 42 Berlin community for making it happen. Want to dive deeper? Explore the Hackathon Wiki and join us in shaping the future—contribute your ideas to the wikis or let us know where these projects should go next in the comments!

‍

‍