Configuration

All course-specific settings are defined in a single CourseConfig dataclass. This is the only thing you need to customize per course — the rest of aita-core is fully generic.

Field Reference

Course Identity

Field	Type	Description
`course_id`	`str`	Short identifier (e.g., `"3102"`). Used in log messages.
`course_name`	`str`	Full display name (e.g., `"CEGE 3102: AI Teaching Assistant"`). Shown as page title.
`course_short_name`	`str`	Sidebar title (e.g., `"CEGE 3102 AITA"`).
`course_description`	`str`	Markdown text shown on the login page below the title.
`system_prompt`	`str`	The LLM system prompt. This is the primary mechanism for pedagogical guardrails.

Week & Topic Mappings

These fields control week-aware behavior: the chatbot won’t discuss topics from future weeks, and content is filtered during RAG retrieval.

Field	Type	Description
`week_topics`	`dict[int, list[str]]`	Maps week number (1-15) to list of topic names covered that week. Used to build the system prompt’s topic awareness.
`topic_num_to_week`	`dict[int, int]`	Maps the number prefix in slide/handout filenames (e.g., `3` from `"3 Topic Name.pdf"`) to the week that topic is first covered.
`hw_num_to_week`	`dict[int, int]`	Maps homework number to the week it’s assigned. Used for week-filtering during retrieval and for the “current homework” injection feature.
`lab_num_to_week`	`dict[int, int]`	Maps lab number to the week it belongs to.
`study_guide_to_week`	`dict[str, int]`	Maps study guide / quiz names to week. Matched by filename prefix.
`example_prompts`	`dict[int, list[str]]`	Example prompts shown as clickable buttons when chat is empty, organized by week. 4 prompts per week works well.

Textbook (Optional)

Field	Type	Description
`textbook_url`	`str`	Base URL for a Wikibook (e.g., `"https://en.wikibooks.org/wiki/Fundamentals_of_Transportation"`). Leave empty to skip textbook ingestion.
`textbook_chapter_to_week`	`dict[str, int]`	Maps chapter URL slugs to week numbers (e.g., `{"Trip_Generation": 2, "Route_Choice": 4}`).

Paths

Field	Type	Description
`base_dir`	`str`	Root directory of the course project.
`course_materials_dir`	`str`	Directory containing course PDFs and LaTeX files.
`faiss_db_dir`	`str`	Where the FAISS vector index is stored.
`docs_dir`	`str`	Where document records (`doc.jsonl`) are saved.
`backup_dir`	`str`	Timestamped backups of previous FAISS indices.
`data_dir`	`str`	Where the SQLite database is stored. In Docker, this is typically a mounted volume (e.g., `/app/data`).

Authentication

Field	Type	Description
`admin_password`	`str`	Password for the admin dashboard.
`cookie_name`	`str`	Name of the JWT auth cookie (e.g., `"aita_3102_auth"`).
`cookie_key`	`str`	Secret key for signing JWT cookies. Set via `GOOGLE_COOKIE_KEY` env var.
`redirect_uri`	`str`	OAuth redirect URI (must match Google Cloud Console config).
`google_client_secret_file`	`str`	Path to the Google OAuth client secret JSON file. Leave empty to disable Google Auth (falls back to student ID login).

LLM & Embedding Settings

Field	Type	Default	Description
`llm_model`	`str`	`"gpt-4o-mini"`	OpenAI model used for chat completions.
`llm_temperature`	`float`	`0`	LLM temperature. 0 gives deterministic, factual responses.
`embedding_model`	`str`	`"text-embedding-3-large"`	OpenAI model for generating embeddings.
`embedding_dimensions`	`int`	`3072`	Embedding vector dimensions.
`chunk_size`	`int`	`2048`	Characters per text chunk during ingestion.
`chunk_overlap`	`int`	`256`	Overlap between consecutive chunks.
`retrieval_k`	`int`	`5`	Number of chunks to retrieve per query.

Config Overrides (Admin Panel)

Settings can be modified at runtime through the Course Settings tab in the admin dashboard. Changes are saved to config_overrides.json in the data_dir and automatically loaded on startup.

Fields that can be edited at runtime:

Course identity (name, description, system prompt)
LLM settings (model, temperature, retrieval_k)
Week schedule and topic mappings
Example prompts
Textbook URL and chapter mappings

Note

Changes to embedding_model, chunk_size, and chunk_overlap require re-running the document ingestion pipeline to take effect.

Computed Properties

CourseConfig provides several computed properties:

google_auth_enabled — True if a Google client secret file is set.
week_to_hw — Inverse of hw_num_to_week: maps week to "HWN" string.
get_topics_covered(week) — Returns list of topics covered up to the given week.
get_topics_not_covered(week) — Returns list of topics not yet covered.