Connecting Models to Knowledge: RAG, Persian Search and Memory
Choose between knowledge files, long context and RAG, build Persian hybrid search, evaluate RAG failures and design agent memory with access control and PII removal.
What you will learn
- A reasoned choice between knowledge files, long context and RAG
- Persian hybrid search and retrieval evaluation
- Secure agent memory with access control
Prerequisites
- AI Inside Workflows
Tools
- ویرایشگر متن
- صفحهگسترده
- Knowledge Files, Long Context or RAG? Chunking and Citations
Choose among simple knowledge files, long context and retrieval-augmented generation; design context-preserving chunks and mechanically auditable evidence citations using local synthetic documents.
- Persian Search: Normalization, Embeddings, BM25 and Reranking
Design Persian-aware hybrid retrieval with tested normalization, lexical BM25, semantic retrieval, rank fusion and reranking; perform an offline exercise without pretending to run an embedding model.
- RAG Failure Modes and Evaluation
Diagnose retrieval and generation failures separately, measure recall and ranking, audit citations and stale evidence, and build an offline evaluation suite without fictitious accuracy guarantees.
- Agent Memory, Access Control, Personal Data and Capstone Project
Design scoped short- and long-term agent memory, enforce consent, per-user authorization, expiry and deletion, and complete an offline capstone testing cross-user isolation and memory poisoning.