refactor: Remove `rubric` and `correct_answer` fields from the `ExerciseResponse` schema and session initialization.

feat: Add and register new learning prompts.

feat: Refactor `/process` endpoint to download files from a provided URL instead of accepting direct file uploads.

feat: Implement a new prompt management system, answer evaluation, and history formatting for learning exercises, and enable Langsmith tracing.

feat: Implement ordering exercise type with item shuffling and extend exercise schemas to include sentence and correct_answer fields.

feat: update exercise generation logic

feat: Provide recent same-concept exercises to the LLM for adaptive generation and cache based on exercise history.

feat: Dynamically generate and shuffle matching exercise right items for API output and update tests accordingly.

refactor: Migrate exercise_type fields and logic to use ExerciseType enum for improved type safety.

feat: Implement a configuration-driven weighted exercise type selection based on Bloom level and mastery.

docs: Add Hugging Face Spaces metadata to README.

refactor: restructure `api/core` into domain-specific subpackages and update imports accordingly.

Add targeted tests for adaptive exercise types

Add adaptive exercise types and matching support

feat: Implement persistent tutor chat history per exercise with robust input sanitization and prompt injection safeguards.

feat: stream adaptive tutor chat responses

refactor: normalize backend exception handling

refactor: sanitize backend status endpoints

refactor: share repository error handling

refactor: harden quiz extract router

refactor: harden pipeline router responses

refactor: harden backend config and error handling

refactor: share tutor chat prompt building

fix: harden quiz tutor api responses

feat: add quiz tutor api endpoint

refactor: unify llm abstractions

refactor: split adaptive prefetch timeout policy

refactor: improve exercise generation logic.

feat: Enhance content processing and generation with improved LaTeX math formatting, Bloom's Taxonomy guidelines, and refined content extraction rules.

refactor: retry adaptive exercise llm calls uniformly

fix: retry transient adaptive llm failures

refactor: introduce content pipeline relation engine

fix: extend graph cycle removal timeout

fix: allow explicit llm timeout overrides

refactor: align subject progress with rl mastery threshold

refactor: remove legacy root pipeline shims

refactor: drop remaining pdf loader env fallback

refactor: remove legacy config shim

docs: add unified content pipeline environment guide

refactor: remove config shim usage from pipeline infrastructure

refactor: unify pipeline settings in backend config

refactor: migrate storage into infrastructure namespace

refactor: migrate processors into infrastructure namespace

refactor: migrate graph into infrastructure namespace

refactor: migrate embed and merge into infrastructure namespace

refactor: migrate llm prompts and utils into infrastructure namespace

refactor: rename subject progress loading helper

chore: clarify subject progress rehydration logs

fix: remove pipeline runtime circular import

fix: tolerate missing underthesea in pipeline startup