The Moscow Transport Department is holding the Moscow Transport Hackathon: from September 25 to October 3, 2026, teams of 3–5 people will solve four production tasks for urban transport, ranging from backup navigation for an autonomous tram without satellite signal to AI forecasting of route load. The prize fund is 4,000,000 rubles, registration is open until September 22 at 23:59 MSK at reg.mt-hackathon.ru, and the authors of strong solutions will have a chance to join the IT team of Moscow Transport.

image
image

What Happened

The Moscow Transport Department announced the Moscow Transport Hackathon, which will take place online from September 25 to October 3, 2026, with the opening and final held offline in Moscow. Registration for teams of 3–5 people is open until September 22 at 23:59 MSK at reg.mt-hackathon.ru. The prize fund is 4,000,000 rubles and is divided equally among four tracks: 1,000,000 rubles per track, of which 500,000 goes to first place, 300,000 to second, and 200,000 to third. The first track is backup odometry for an autonomous tram without GNSS: a mathematical model and ROS 2 Humble nodes in C++ or Python are required. The second is AI forecasting of tram route load based on validation data with horizons of 1 day, 1 month, and 1 year on Java 17+ with Spring Boot and React. The third is a predictor of transport delays 10–15 minutes before the event based on NDTP streaming telemetry using CatBoost or PyTorch. The fourth is a gamified simulator for high-speed rail staff, designed for speeds up to 400 km/h. The announcement was confirmed by media publications on August 31, 2026.

Context

Formally, this is not a technology release, but a procurement of ideas through a competition, and the problem statement itself reveals the client's internal workings. Data for load forecasting comes from the Unified Data Center of the State Unitary Enterprise "Moscow Metro", and the delay predictor is built on NDTP streaming telemetry, meaning participants work with the same types of data that power the production environment of Moscow transport. Odometry without GNSS is the only track with an explicit research component: navigation without satellite signal usually requires sensor fusion, i.e., a combination of IMU, wheel odometry, and map matching, and the requirement for a mathematical model and ROS 2 Humble nodes sets a reproducible engineering artifact. Load forecasting with horizons of 1 day, 1 month, and 1 year is effectively three different tasks: the daily horizon is a standard ML case based on validation data, while the annual horizon is statistically weakly validated within a hackathon and is closer to planning than to prediction. A delay predictor 10–15 minutes ahead is a classic early warning task, and such a horizon is operationally meaningful for dispatcher decisions.

Why This Matters for the Industry

Urban transport authorities are turning to external IT teams as a source of solutions in four specific niches: backup navigation without satellite signal, ML forecasting of passenger flow, proactive prediction of delays based on streaming telemetry, and simulation training for high-speed rail staff. The client openly publishes the production stack and data types — ROS 2 Humble, CatBoost and PyTorch, Java 17+ with Spring Boot and React, NDTP telemetry and validation data — which is a rare public cross-section of the real technological landscape of Moscow transport IT and a hint at which skills are in demand there. For startups and studios, the published tasks serve as free demand research: the cheapest way to test a niche, find engineers, and build relationships with the IT team of Moscow Transport is to enter a track where there is already some groundwork. If the practice of open cases on urban data becomes established, the industry may gain a talent pipeline, and with the disclosure of datasets, de facto benchmarks for urban transport in the Russian Federation, but this is an interpretation, not a client commitment. Over the next couple of years, the model of "hackathon as a simplified procurement channel" is also being discussed, from which niche vendors for backup navigation, predictive analytics, and training simulators may emerge; for now, this is a speculative part.

Why This Matters for Users

For the reader, this is a competition with verified dates and money, and the main action is one: manage to assemble a team of 3–5 people and submit an application by September 22 at 23:59 MSK at reg.mt-hackathon.ru. It makes sense to assemble a team for specific roles — ROS 2 developer, ML engineer, backend in Java or Python, frontend in React, game designer — and choose a track based on your stack, not the other way around. The tasks are solved on real urban data such as validations and NDTP telemetry, so even without a prize, this is a ready-made case for a metropolis-level portfolio, which is difficult to obtain otherwise. A path to the IT team of Moscow Transport is announced for the winners, and the window until the start on September 25 is short enough to plan data work in advance.

What Is Still Unknown / Limitations

The evaluation methodology, metrics, and dataset composition have not been published, so it is difficult to compare the quality of solutions in advance. In the delay predictor track, the definition of the event, i.e., the method of building labels, and the validation scheme are not disclosed: without them, even a strong baseline cannot be honestly compared with other solutions, and the clarity of the pipeline is not equal to the verifiability of the result; the value of the 10–15 minute horizon depends on unpublished metrics such as precision and recall by delay types and the cost of false positives. The criteria for winners to join the IT team of Moscow Transport are not disclosed, nor are the plans for disclosing datasets and repeating the format, so scenarios with benchmarks and "hackathon as a procurement channel" remain interpretations.

Sources

Author

Look at AI, editorial team