Inter RAO, with Yandex support, launched the "AI Assistants for Energy" competition on the Codenrock platform: participants use prepared data from one of Russia's largest energy companies to build AI assistants for office processes — from corporate presentations and court documents to IT/cybersecurity digests from news and BPMN 2.0 diagrams. Team registration is open until September 27, 2026, the in-person final in Moscow will take place on November 3, the prize pool is 750,000 rubles, and the organizers promise the best projects a chance to enter Inter RAO's production ecosystem.

image

What happened

The competition is open to teams of one to five people; the recommended composition is a developer, ML specialist, analyst, and designer. Four tracks are announced: a generator of corporate presentations from a text brief, recognition and classification of court documents with entry into a database, an assistant for IT/cybersecurity digests from a news feed, and a generator of valid BPMN 2.0 diagrams from a text description. Registration will continue until September 27, the start is scheduled for September 28, from October 9 to 23 the strongest ideas will be refined together with the client's experts, and on November 3 in Moscow the finalists will present their projects in person. According to the organizers, the best projects may enter the Inter RAO ecosystem.

Context

The post about the competition is an advertisement: the link contains an erid marker, and on Codenrock and in event aggregators the event is positioned as a hackathon. There is no scientific novelty in it: no new models, methodologies, or benchmarks are described in the sources, and technologically nothing will change in the coming weeks — this is a corporate competition, not research. The valuable signal here is different: a major energy customer is opening prepared data and expert mentorship, thereby showing which AI tasks it considers a priority. All four tracks can be solved with existing LLMs plus engineering work — structured output, validation, and post-processing, so the barrier to entry for participants is low, and the focus shifts from the models themselves to the reliability of pipelines over real document flows. The most verifiable of the four tasks is the generation of valid BPMN 2.0 diagrams: the validity of the schema can be objectively checked against the specification, whereas the quality of presentations or digests is not measured so clearly.

Why this matters for the industry

For the industry, this is another signal that corporate demand is shifting from RAG chatbot pilots to agentic pipelines over real document flows: a major energy customer is testing generative AI on end-to-end office processes — presentations, court documents, news monitoring, and process diagrams. The format is also important: the customer itself opens data and mentorship, and the best solutions risk not remaining a demo but entering Inter RAO's production ecosystem. For AI teams and integrators, such a competition is a cheap distribution channel: a month of work with prepared data from a major client, mentorship, and a stated path to production without lengthy corporate procurement. If the organizers' promise is at least partially fulfilled, the market will have a reproducible template of "corporate competition, mentorship, implementation," and the energy sector may become one of the anchor customers for document AI in Russian; this is a forecast, not an established fact.

Why this matters for users

For readers, this is a free opportunity to bring an LLM prototype — presentation generation, document AI for court documents, news digests, or text-to-BPMN 2.0 — to a production case with customer data in a month. To try it, you need to assemble a team of one to five people (recommended composition — developer, ML specialist, analyst, and designer), choose one of the four tracks, and register by September 27, 2026, at the link https://codenrock.com/contests/ii-assistenty. Refining strong ideas with experts from October 9 to 23 is essentially a free product sprint, and participation gives a strong line in the portfolio even without a win. Winners receive prizes from the 750,000-ruble fund and a chance to see their solution in the ecosystem of a major energy company.

What is still unknown / limitations

The competition methodology is opaque: only "prepared data" is announced, and the evaluation criteria, jury composition, allowed models, and BPMN validity requirements are not disclosed, so it is impossible to assess in advance the quality of solution checking and their reproducibility. The four tracks are task names from an advertisement, not ready-made specifications for agentic pipelines: there are no quality metrics, acceptance thresholds, or test sets in the sources. The statement about the best projects entering the Inter RAO ecosystem is a promise from the organizers, not a confirmed fact; the answer will only appear after the November 3 final. In addition, on ICT2GO the event is listed with dates of September 28–30, 2026, which conflicts with the announced schedule where the in-person defense is scheduled for November 3, so key dates should be rechecked on the official competition page.

Sources

Author

Look at AI, editorial team