Your next data challenge is just a few steps away.

At ML Challenges, we organize AI competitions and benchmarks.

AVALIA

Our Next Generation AI Competitions and Benchmarks Platform — under development

About ML Challenges

We build the competitions and benchmarks that put machine learning methods to the test. A good challenge turns a vague question into a measurable one: a dataset, a metric, a leaderboard, and a protocol everyone involved can trust.

Getting there takes more than a platform. It takes a problem worth solving, an evaluation that cannot be gamed, and infrastructure that holds up when hundreds of submissions land at once. We handle that part, from the first conversation to the results paper, for research labs, industry teams and conference competition tracks.

11Projects Completed

€1M+Total Prize Pool

200+Organizers Supported

5+Countries Reached

About the Founder

Adrien Pavão

Founder, Benchmark Scientist

Our Services

Competition

Organization of a competition, with deadline and prizes

Benchmark

Setup of a benchmark, either public or private

Full Platform

Deployment of your own server of the platform Codabench

Technical Setup

Implementation and set up of your challenge

Scientific Expertise

Scientific expertise to support the design and analysis of your challenge

Confidential Data

Our protocol allow you to organize competitions and benchmarks on confidential data (medical, industrial)

All Services

Featured Projects

Competition·2025

Survival AnalysisFinanceBenchmark

FinSurvival Challenge

Advancing survival modeling for financial transactions

Survival modeling is the statistical prediction of the time until a specific event occurs, and it holds immense value across the financial sector. It is used to estimate the time until loan repayment or default and to predict portfolio churn, which gives critical insights for risk management, strategic planning and customer retention. Winning teams presented their solutions at the ICAIF'25 conference.

With ICAIF'25 Conference · ChaLearn · Ava Labs · Rensselaer Polytechnic Institute

$1,750 prize poolWebsite Competition

Competition·2024

Cooperative AIAgentic AILLM Agents

NeurIPS Concordia Contest

Cooperative intelligence for AI agents

A NeurIPS 2024 competition in the field of cooperative AI, focused on the cooperative intelligence of AI systems, that is, the ability of an agent to achieve its goals in ways that also promote social welfare, across diverse environments and partners. The scenarios test skills such as promise-keeping, negotiation, reciprocity, reputation, partner choice, compromise and sanctioning.

With Google DeepMind · Cooperative AI Foundation

$10,000 prize poolHomepage Competition

HackathonUpcoming

EnergyHackathon

EDF Hackathon 2026

A new edition of the EDF hackathon, following the 2025 event. Topic, dataset and registration details will be announced soon.

With EDF (Électricité de France)

Platform·Ongoing

Open SourceBenchmarksCompetitions

Codabench

The open-source platform for AI benchmarks and competitions

Codabench is a web-based platform for organizing scientific challenges and benchmarks in artificial intelligence, where algorithms compete on tasks in fields like medicine, physics, linguistics, economics and ecology. We are core contributors to the project, and we deploy and operate dedicated instances for our clients.

With ChaLearn · LISN, Université Paris-Saclay

30,000 users · 500 public competitionsPlatform Source code Documentation White paper

Our role

Codabench is developed and hosted primarily in France 🇫🇷, at LISN, Université Paris-Saclay, together with the non-profit ChaLearn. Our team has been part of the core contributors throughout the maturation of CodaLab Competitions and then Codabench: platform development, release management, community support, and the design of the evaluation protocols that organizers rely on.

We also deploy and operate dedicated instances of the platform for clients who need their own server, including setups where the evaluation data never leaves the organizer's machines.

Key features

Open source: the entire platform code is open source and welcomes contributions, supporting transparent AI evaluation.
Containerized environments: Docker and Podman isolate test environments, ensuring reproducibility and simplifying logistics.
Decentralized computing: organizers can connect their own compute resources to challenges, making the system fully scalable.
Flexibility: organizers provide their own evaluation code, enabling virtually any experimental protocol.
Confidential data: organizers retain full control over evaluation data, which never has to be uploaded to the main server.

Impact

Codabench and its predecessor CodaLab rank among the largest global actors in AI competitions, hosting one of the highest numbers of competitions worldwide. Two years after its official launch in August 2023, the main server counts:

30,000 users
500 public competitions
Over 200,000 algorithm submissions from participants

All Projects

Testimonials

"We have successfully set a complete dedicated platform environment with Adrien for an industrial project with highly confidential data that noticeably allowed improvements to the open-source project. I have also taken part to a workhop session lead by Ihsan about deploying competitions on Codabench. It is a pleasure to work with their team: from designing innovative problem statements to ensuring seamless execution, they demonstrate high mastery. More precisely, they can assemble successfully all competition core components from the dataset and the baseline to the workflow of technical constraints and organizers' wishes, including configuring the competition."

Anne-Catherine Letournel

Research Engineer @ LISN

"Adrien nous a appuyé pour l'organisation d'un challenge interne au groupe dans le domaine de la prévision de séries temporelles. Son expertise de l'outil Codabench mais aussi sa connaissance de ce type d'évènement nous a été précieuse. Le sujet était complexe et faisait appel à des fonctionnalités avancées de la plateforme. L'évènement se déroulait sur une journée, avec des soumissions possibles en R et en Python, induisant des contraintes en termes de dimensionnement de la charge (de nombreuses soumissions de code en parallèle). Merci encore Adrien pour ton support et, ce qui ne gâche rien, ta bonne humeur tout au long de l'aventure, ta réactivité et ta disponibilité !"

Yannig Goude

Senior Researcher @ EDF

"Adrien has worked with us on several iterations of the « Learning to run a power network » (L2RPN) challenge series along the years, helping on its development and deployment on Codabench. The smooth execution of the competitions on Codabench has been key for the worldwide success of this Grand Challenge. Those competitions came in particular with advanced setup for running artificial agents, often based on reinforcement learning, to operate power grids. Some competition even required model retraining on the competition platform for more advanced agent evaluation. Adrien has demonstrated great expertise on this matter overall. He has further taken leadership on the DMLR Book writing on AI Competitions with several co-authors including me. He is definitely someone who can help you run successful competitions and benchmarks when you are looking for support."

Antoine Marot

AI Project Lead @ RTE

Trusted By

Ready to launch your challenge?

Let's discuss your project and find the best approach together.

Get in Touch