Build an agent that answers questions about kitchen CCTV.

Build an agent we can run again. It receives kitchen videos and a JSON file of questions. It returns an answer for each question, the supporting timestamps or frames, and a run log. Nobody may inspect the videos manually during scoring.

python answer.py --videos ./videos --questions questions.json --out answers.json --log run_log.json

Find the moments that answer the question. Return not_visible when the footage does not show enough. The example command above is the preferred interface; the full specification explains the required files.

Provide your repository or endpoint URL, agent name, models or APIs used, expected cost per scored run, and any setup notes. Your agent must produce answers, supporting timestamps or frames, and a run log.

The published deadline has passed. Check entry status with submit@builderr.ai first. The full specification explains the required files.

We run your submitted agent on kitchen video with questions you have not seen before, then compare its answers with the answer key. It must find supporting moments in the footage without manual inspection during scoring.

For example, a question may ask when a handoff happened. Return the timestamp and supporting video evidence; if the footage does not show enough, return not_visible.

Total score is 100 points. A run is valid only if it stays under the hard cost and runtime caps. If it goes over, it does not rank.

• 80 points — correct answers: hidden yes/no, counts, event order, timestamp, duration, and not-visible questions.
• 15 points — lower model/API cost: lower model/API cost helps only after the answer is valid and useful.
• 5 points — a run we can repeat: one command, useful evidence, frame/model call log, and repeatable outputs.

Hard cap: $0.30 estimated model/API cost per 60 minutes of source video, 25 minutes wall-clock, and about 1,500 sampled frames per 60 minutes unless the evaluator states an equivalent frame budget.

If the scored set is longer or shorter, the dollar and frame caps scale with video minutes. The cap is deliberate: a useful restaurant tool has to find the right windows cheaply, not send every frame to the biggest model.

Start with the long fixed-view kitchen clips below. Scoring uses hidden questions and an answer key on the stated video set; if any held-back clips are added later, that will be stated before they are used.

These are examples so builders understand the job. Final scoring uses hidden questions on the same public videos, with an answer key.

Open-source or local models are recommended, because this should be cheap enough for a small kitchen to run often. They are not mandatory.

Cloud models are allowed if every call is logged and the normalized cost stays under the cap. Local runs count as $0 model/API cost because they use no paid model/API calls. This excludes local hardware and electricity costs. They still have to fit the same runtime and frame limits.

The baseline is a simple coarse-to-fine video agent: sample the whole clip cheaply, identify likely time windows, inspect only those windows with a vision model and OCR where useful, then answer in JSON with evidence.

A result shows how an entry answered the tested kitchen questions and what its model/API calls cost. The current six-question check is limited evidence; it does not establish accuracy across every kitchen, camera or operating condition. Use the linked videos and run evidence to understand that scope.