MODEL_OPS_AND_APP_SPLIT.md

Chatbot App Split and Service-Intent Model Ops

The chatbot project now keeps responsibilities separated without moving internal orchestration into Django.

Apps

• apps.chat

  • Chat API/UI.
  • Knowledge-base retrieval.
  • Runtime service detection.
  • internal orchestration microservice execution client.

• apps.service_intents

  • Dataset ownership commands.
  • Rebuilds runtime and model-training service intent CSV files.

• apps.model_ops

  • Optional Hugging Face model installation.
  • Fine-tuning service-intent/slot extractor.
  • Optional ONNX dynamic quantization.

Datasets

• apps/chat/training_models/data/service_intents.csv

  • Runtime detector only.
  • Original six columns.

• apps/chat/training_models/data/service_intents_model_training.csv

  • Model training dataset.
  • Includes is_service, parameters_json, missing_fields_json, role_context, service_action, and source.

Commands

Rebuild both datasets:

python manage.py build_service_intent_datasets

Install optional ML dependencies:

python -m pip install -r requirements-ml.txt

Install the base model, train, and quantize:

python manage.py prepare_service_intent_model --rebuild-datasets

Dry-run without downloading/training:

python manage.py prepare_service_intent_model --dry-run --skip-install

internal orchestration

internal orchestration remains an external microservice. Django still talks to it through the existing service executor.