In-Depth Criteria and Tips

MAIN CRITERIA: Your project has to be built on Mistral's APIs, not other model providers. Projects that do are ineligible for prizes no matter how well they score. Feel free to ask staff if you're unsure what is fine and what isn't.

Value + Human Insight (0–15)

Whether you understand the person you're building for, whether what you've made would genuinely matter to them, and whether it answers the problem statement directly rather than an easier one sitting next to it.

Tips: Build for a specific person and the counterpart they need to reach (e.g. the designer and the ops lead), not a generic "everyone." Teams are cross-disciplinary, so the strongest projects help someone in one discipline be understood by, or make sense of, someone in another. Back the need with evidence, whether that's user conversations, data or lived experience. We should be able to picture both people/teams reaching for this.

Technical Execution (0–15)

The depth of what you actually built. Is it more than something you could vibe-code in a couple of prompts?

Tips: Show us real engineering: meaningful integrations, non-trivial logic, a hard problem genuinely tackled, working reliably by demo time. Favour specialised APIs and real workflows over one giant prompt. Chain models or agents so the pipeline degrades gracefully, pick the right model per step for size, latency and cost, and use multimodal where it helps you read documents, images or recordings. If your system interprets or condenses, show it can be trusted: provenance to source, visible uncertainty, tested on messy real input. Making sense of information is only one hard problem here. Presenting an idea across disciplines, holding a decision and its reasoning together, or structuring a review count just as much. Production polish isn't expected. Evidence of a hard problem solved is.

Creativity + Design (0–10)

How novel your idea is and how well it's designed. Are you doing something fresh, and is it genuinely pleasant to use?

Tips: Push past the obvious first idea, then pair it with design good enough that concept and craft reinforce each other. Polish here means interface craft and coherence, not production readiness. Reliability is judged under Technical Execution. Have a visual idea before you start prompting the frontend, and make sure it holds up across every input you claim to support. We're hoping for a "why didn't this exist already?" reaction.

Feasibility + Scalability (0–10)

Whether it could work in the real world now, and grow from there across other domains, contexts and teams rather than just the one you demoed.

Tips: Have credible answers for the real technical, legal and social barriers, including the privacy, security and safety of what you process. Better still if your design improves on the status quo rather than just avoiding harm. Generality matters a lot to us: a solution wired to one data type from one team scores below one whose approach carries across. Prove it for a specific example, but the mechanisms should be general. Demo one team's real workflow, and show us it transfers without a rebuild.