How supoz9/edge-ai-education works
Edge-AI-Education is a local Bac Pro CIEL tutoring system: a student asks through Open WebUI, the API retrieves permitted course passages, builds a Socratic prompt, and requests an answer from Ollama. Its distinct content stages are corpus organization, ingestion and chunking, hybrid retrieval, and guided generation. The graph includes the documented restricted-solution safeguard and historical agent profiles. Ingestion, retrieval, and API wiring are supported by sampled source; Open WebUI and deployment details come from the README and tree, so uncertain runtime links are omitted.
Main entry points: ingest.py Corpus ingestion · api.py Chat API · retrieval.py Hybrid retrieval · 01_referentiel Referentiel
Components 17
Learning corpus
- Referentiel
corpus_ciel/01_referentiel - Courses
corpus_ciel/02_cours - TP statements
corpus_ciel/03_tp_enonces - Datasheets
corpus_ciel/04_datasheets - Restricted solutions
corpus_ciel/05_corriges - Restricted hints
corpus_ciel/06_coups_de_pouce
RAG pipeline
- Corpus ingestion
rag/ingest.py - Metadata and chunks
rag/chunking.py - Hybrid retrieval
rag/retrieval.py - Chroma index
- BGE-M3 embedder
Tutoring runtime
- Open WebUI
- Chat API
api-openwebui/api.py - Socratic prompts
api-openwebui/prompts.py - Ollama
Agent profiles
- Demo tutors
agents
Other components
- Student
Connections 17
- Student → Open WebUI: asks through
- Open WebUI → Chat API: sends chat
- Chat API → Hybrid retrieval: requests passages
- Chat API → Socratic prompts: builds messages
- Chat API → Ollama: requests generation
- Chat API → Open WebUI: returns answer
- Referentiel → Corpus ingestion: read
- Courses → Corpus ingestion: read
- TP statements → Corpus ingestion: read
- Datasheets → Corpus ingestion: read
- Restricted solutions → Corpus ingestion: excluded by default
- Restricted hints → Corpus ingestion: excluded by default
- Corpus ingestion → Metadata and chunks: parses and chunks
- Corpus ingestion → BGE-M3 embedder: encodes chunks
- Corpus ingestion → Chroma index: writes vectors
- Hybrid retrieval → BGE-M3 embedder: encodes query
- Hybrid retrieval → Chroma index: queries index