CinvanaAI
← All projects

Experiment · Workflows & capabilities / Code & architecture

AI Software Factory Prototype

Follow a two-stage code-generation experiment from the first draft to a separate refinement.

Repository: CinvanaAI/ai-software-factory-prototype

Recorded synthetic example. An executable experiment using supplied synthetic inputs.

See the idea in action.

Separate engineer-output, final-output and a job-evidence.json record.

Example input
One software task plus injectable engineer/refiner functions.
Captured result
{
  "task": "Build a greeting tool",
  "status": "complete",
  "stages": [
    {
      "stage": "engineer",
      "input_files": 0,
      "output_files": 2
    },
    {
      "stage": "refiner",
      "input_files": 2,
      "output_files": 2
    }
  ],
  "engineer_files": {
    "app.py": "def main() -> None:\n    print('Build a greeting tool')\n\n\nif __name__ == '__main__':\n    main()\n",
    "README.md": "# Generated scaffold\n\nTask: Build a greeting tool\n"
  },
  "final_files": {
    "app.py": "def main() -> None:\n    print('Build a greeting tool')\n\n\nif __name__ == '__main__':\n    main()\n\n# Reviewed by the simulated refiner.\n",
    "README.md": "# Generated scaffold\n\nTask: Build a greeting tool\n"
  },
  "failed_refiner": {
    "error": "Synthetic refiner failure",
    "engineer_output_retained": true,
    "completed_job_record_exists": false
  },
  "generated_code_executed": false
}

Try the example.

From the repository root, follow the dependency requirements in the README. This example uses supplied synthetic material.

python -m pip install -e .
python -m software_factory "Build a greeting tool" --workspace ./demo-output

Separate engineer-output, final-output and a job-evidence.json record.

Complete setup and instructions ↗

The interesting part.

Keeping stage outputs separate makes a generation pipeline inspectable even when a later stage fails.

Public continuation

Where it came from.

Factory experiments with simulated scaffold/refine stages and broader orchestration trials.

Turns the authored two-stage idea into a contained offline proof with evidence; excludes copied third-party implementations.

The public code does not supply GPT-Engineer, Aider or Docker integrations.

Source ↗

Follow the family: Builder, Factory and the Python Foundry