dmcc

DeepMind Code Contests (DMCC)

An environment for solving DeepMind Code Contests problems in Python and C++.

Running this example

Prerequisites:

• Docker
• taiga-local-tunnel installed and logged in (from the repo root: uv pip install -e local-tunnel/ && taiga-local-tunnel login)

Build the environment:

cd examples/dmcc
docker build --build-context taiga=../../ --platform linux/amd64 -t dmcc -f Dockerfile .

Then start an interactive session against the hosted Taiga platform:

taiga-local-tunnel start \
  --image dmcc \
  --metadata-file problems-metadata.json \
  --problem-id python1575a

This will open a browser where you can watch Claude solve the python1575a problem. The container is cleaned up automatically when you press Ctrl+C.

Generating the dataset

The per-problem test data is not baked into the image — it's attached to each problem via preloaded_files and the container reads it from extra_fields.data_file at runtime. Generate the files locally with:

cd examples/dmcc
uv pip install '.[dev]'
python prepare_dataset.py

This downloads the DeepMind Code Contests dataset and writes one JSON file per problem to data/problems/ (gitignored). problems-metadata.json references those paths under preloaded_files; Taiga uploads them when you create the problem.

Available problems

• python1575a - Problem 1575_A (Another Sorting Problem) in Python
• cpp1580e - Problem 1580_E in C++

To run a different problem, change --problem-id python1575a in the command above.

Problems are from the DeepMind Code Contests dataset.