x3allamerican/ai-program-skills/rag-over-course-content

RAG Over Course Content

RAG (Retrieval-Augmented Generation): index course content → search → use as context → generate. Course-aware AI.

RAG architecture

1. Embed course content — turn into vectors
2. Vector database (Pinecone, Weaviate, Chroma)
3. Student question — convert to vector
4. Similarity search — find relevant content
5. Pass to LLM — context for response
6. LLM generates — informed by retrieved content

Implementation steps

1. Chunk course content — into 200-500 token segments
2. Embed each chunk — using embeddings model (text-embedding-3-small)
3. Store in vector DB
4. Build query function — embed query, search
5. Inject context into prompt — top 3-5 chunks
6. Generate response — using LLM

Quality factors

1. Chunk size — too small loses context; too large loses precision
2. Embedding model quality — varies
3. Search relevance — return truly related content
4. Citation in response — point to source
5. Refresh frequency — when content updates

Where this fits in the X3 empire

RAG implementation for CardPrepAI Academy AI assistant.