Skip to content

TeacherPro v2.2.0

Choose a tag to compare

View all tags
@github-actions github-actions released this 26 Jun 15:23
· 20 commits to main since this release
v2.2.0
1055c9a

TeacherPro v2.2.0 – Knowledge Database & RAG

New Features

📚 Wissensdatenbank (Knowledge Database)

  • Subject-based vector database for teaching materials (PDFs)
  • Create subjects, grade levels, and topics directly in the app UI
  • Import PDFs via native file dialog → automatic text extraction + chunking
  • Embedding via bge-m3 (multilingual, 100+ languages, 1024-dim vectors)
  • Cosine-similarity search across all embedded chunks

🤖 AI Chat with RAG

  • Retrieval-Augmented Generation: AI chat automatically searches the knowledge database when a subject/topic is selected
  • Hierarchical database selector in chat panel (collapsible tree)
  • Retrieved content is injected as structured context into the AI prompt
  • Native Chat Messages API (ai_generate_chat) → proper message arrays instead of concatenated strings
  • Token Context Bar: Visual indicator showing lesson/history/RAG/system token usage
  • Dynamic num_ctx / num_predict adjustment when RAG or thinking mode is active

📄 PDF Processing

  • Digital PDF text extraction (via pdf-extract, pure Rust, cross-platform)
  • Scanned PDF OCR fallback (via system tesseract + pdftoppm if installed)
  • Intelligent chunking with UTF-8 safety, paragraph-aware splits, configurable overlap

🎨 UI Improvements

  • SubjectDbManager: Full tree UI for managing database folders and files
  • ChatContextBar: Token budget visualization with color-coded segments
  • AiMarkdown: Now renders markdown tables as HTML <table> elements
  • Custom dropdown: Dark-mode-safe database selector replaces native <select>
  • Delete individual files: Remove specific PDFs from the database
  • Progress events during import/embedding pipeline

Technical Improvements

  • Refactored Rust backend with subject_db module (pdf, chunk, embed, store)
  • All database operations async via Tauri commands
  • Fully cross-platform: Windows, macOS, Linux
  • Ollama embeddings (bge-m3) run on Apple Silicon, NVIDIA CUDA, AMD ROCm
  • Auto-pull of embedding model on first use

Breaking Changes

  • Minimum Ollama version: 0.5+ (for /api/embed endpoint)
  • bge-m3 model (~2.2 GB) required for embedding; auto-pulled on first import

Files Changed

  • 12 modified files, 12 new files (including tests)
  • Core Rust: subject_db/ module (5 files), lib.rs
  • Core Frontend: Editor.tsx, store.ts, ChatContextBar.tsx, SubjectDbManager.tsx, SettingsModal.tsx
  • i18n: de.ts, en.ts (full translations)
Assets 9
Loading