Dialog — Diagnostic Interactive Assessment of Learning through Open Grading

Emergency Nursing Course Processor — transforms raw nursing course material (PDF/text) into categorized knowledge chunks and test questions via a LangGraph pipeline.

Quick Start

cp .env.example .env          # fill in your Ollama Cloud key
docker compose up --build

The processor API will be available at http://localhost:8000.

Mock Mode (no LLM tokens)

MOCK_LLM=true docker compose up --build

Endpoints

Method	Path	Description
GET	/health	Health check
POST	/process	Upload a PDF/TXT and run the pipeline

Local Development

uv sync
MOCK_LLM=true uv run python main.py api

Or process a single file:

uv run python main.py docs/nursing_sepsis_learning_module.pdf

Tests

uv sync --all-extras
uv run pytest tests/ -v

Project Structure

dialog/                          # repo root
├── pyproject.toml               # single project file
├── main.py                      # CLI entry point (api / file processing)
├── Dockerfile
├── docker-compose.yml
├── docs/                        # sample course material
├── tests/                       # all tests
│
└── dialog/                      # installable package
    ├── default_config.py        # config dict + env-var overlay
    ├── api.py                   # FastAPI endpoints
    ├── agents/                  # agent factories grouped by role
    │   ├── schemas.py           # Pydantic structured-output models
    │   ├── chunker/             # create_semantic_chunker(llm)
    │   ├── questioner/          # create_question_generator(llm)
    │   ├── auditor/             # create_quality_auditor(llm)
    │   ├── classifier/          # create_dept_classifier(llm) (deferred)
    │   └── utils/               # AgentState, shared helpers
    ├── graph/                   # graph orchestration (no agent logic)
    │   ├── processor_graph.py   # CourseProcessorGraph orchestrator
    │   ├── setup.py             # node/edge wiring
    │   ├── propagation.py       # initial state creation
    │   └── conditional_logic.py # routing (future: human-in-the-loop)
    ├── dataflows/               # data-source abstraction
    │   ├── pdf_parser.py        # PyMuPDF
    │   └── text_parser.py       # plain text / markdown
    └── llm_clients/             # LLM provider abstraction
        ├── base_client.py       # ABC
        ├── factory.py           # create_llm_client()
        ├── openai_client.py     # Ollama / OpenAI compat
        └── mock_client.py       # FakeListChatModel for testing

Environment Variables

Variable	Default	Description
OLLAMA_API_KEY	—	Required. API key from ollama.com/settings/keys
OLLAMA_BASE_URL	https://ollama.com	Ollama Cloud endpoint
OLLAMA_MODEL	gemma4:31b-cloud	Chat model
MOCK_LLM	false	Run pipeline with deterministic mock responses