PokeRAG: Build Your Team, Enter the Arena

Neovision & NSIGMA AI Hackathon — April 2025

AI: A Serious Playground

When NSIGMA, the Junior Enterprise of Ensimag, decided to organize an AI hackathon open to all students, they turned to Neovision, their official partner. The brief: a topic that is educational, innovative, and engaging.

Neovision designed a challenge around one of the hottest topics in AI: RAG. A cutting-edge, concrete theme, buzzing across the professional world. We guided the teams, answered their technical questions, challenged their architectural choices, and evaluated their solutions with both rigor and kindness. NSIGMA, on their side, managed all campus logistics with the energy of a speedrun. NVIDIA, our partner for the event, offered every participant an online AI course — a nice bonus to extend the experience beyond the hackathon.

The RAG Challenge: When AI Enters the Pokémon Arena

For this edition, we designed a challenge around a RAG (Retrieval-Augmented Generation) pipeline, using a database well known to students: Poképedia. The goal: build a conversational agent capable of answering all kinds of questions about the Pokémon universe, relying on a subset of documents from the encyclopedia.

Behind this playful appearance hid a true technical challenge. The goal wasn’t to build a model from scratch, but to design a functional RAG pipeline capable of answering precise questions about the Pokémon universe. Each team therefore had to:

· build a solid RAG pipeline as sturdy as a Snorlax, capable of dynamically querying Poképedia;

· set up a search strategy as sharp as a Pokédex, with vectorization, indexing, and similarity calculation;

· connect everything to an LLM to generate answers as accurate as a well‑placed Thunder Shock.

It was a full learning arena, allowing students to develop concrete skills while having fun taming a knowledge base as rich as it is varied.

Communication That Hit the Mark, Like a Pokéball Dead Center

LinkedIn to attract curious trainers, Discord to coordinate teams like in a Pokémon Center, TV screens and posters to cover the campus: the communication strategy was right on target.

In less than two weeks, around twenty future AI developer talents joined the league. A real critical capture on the first throw!

The Trainer Timeline of This AI Hackathon

Two Days to Catch All… the Score Percentages!

Day 1

Right from the first evening, things got serious. Participants faced a dual challenge: quickly grasp a dense technical topic while organizing themselves efficiently as a team. Brains heated up fast, early diagrams filled whiteboards, and keyboards began to sizzle.

One important point of attention was highlighted: the parsing phase. Cleanly extracting raw text from PDF files is far from trivial, especially knowing that the most robust methods can take several hours to run… and that the available machines (PCs without GPUs) were not exactly computational beasts. Planning, anticipating, and sometimes… letting scripts run overnight was essential. Welcome to the life of an enthusiast!

The second evening of this AI hackathon marked a turning point: that of optimization. Most teams had a functional first version by the start of the evening. The goal? Find improvement levers to climb the leaderboard.

But there was one last twist: results could only be submitted for evaluation every 15 minutes. The result? An Olympic‑final atmosphere at every “push”: eyes locked on the clock, coffee in hand, intense discussions to the second… each attempt could change the ranking. A true algorithmic thriller.

Day 2

Diplomas and Trophies for the Best Data Trainers

After two days of intense competition, technical exploration, and shared good vibes, the long‑awaited moment arrived: the award ceremony.

Lucas, our CEO, opened the ceremony by reminding everyone of the goals of this hackathon: manipulating cutting‑edge AI tools, understanding the mechanics of a RAG pipeline, but also experimenting as a team, adapting, pivoting, collaborating.

Guided by experts, students got the chance to tackle a concrete challenge — an experience both demanding and rewarding, and a strong stepping stone for the rest of their journey.

Then, in an atmosphere charged with excitement — worthy of a decisive match in a Pokémon arena — the podium was revealed. Three winning teams will have the opportunity to join us in our offices for an exclusive AI training session:

A huge congratulations to all participants: you did far more than code. You dared, built, innovated. And above all, you proved that AI can be as playful as it is demanding.

The award ceremony continued around a lively buffet, filled with experience sharing, technical anecdotes, and great encounters. A beautiful conclusion to these two intense days!

Crossed Perspectives and XP Sharing on This Adventure

“Designing a technical and challenging task while making it interesting and achievable for students in two four‑hour sessions was fascinating. Just a few months after my graduation, returning to my school to supervise a hackathon — this time as an organizer rather than a participant — was a particularly rewarding experience. The enthusiastic feedback we received during the event and in the following days confirmed the relevance of this initiative. Can’t wait to do it again!”

Loris, R&D Engineer specializing in RAG and organizer of this hackathon

“We organized this AI hackathon to help students grow their skills, push beyond their limits, and showcase their talents in AI, coding, teamwork, creativity, and stress management.
Participants really enjoyed this experience, especially since many of them had never taken part in a hackathon — let alone on a subject we’re all passionate about.
I give this Hackathon a solid 10/10 and I’ll end with: Neovision X Nsigma best partnership!”

Hiba, Secretary General of the NSIGMA Junior Enterprise at ENSIMAG