# AetherMind — What This Model Knows

AetherMind is a local RAG assistant. It answers ONLY from a curated knowledge
base of PDFs about Retrieval-Augmented Generation (RAG) and AI agents. It does
not browse the internet, so questions outside this scope will be answered with
"that is not in my knowledge base."

## Knowledge areas
1. RAG fundamentals — when to use RAG vs fine-tuning, failure modes.
2. Vector databases & retrieval — embeddings, chunking, vectorless / page-index retrieval, TurboVec.
3. Agentic RAG, agent memory & MCP — retrieval decisions, cross-turn memory, tool/DB access via MCP.
4. Production RAG — architecture blueprints, cost/latency, evaluation, security guardrails.

## Example questions (use-case style, not generic)
- When should I use RAG instead of fine-tuning, and what are the trade-offs?
- Walk through a real production example: how would KARLA-style retrieval be wired into a support bot, and where does it get used?
- What chunk size and overlap give the best retrieval for long technical PDFs, and why?
- Compare vectorless RAG with page-index retrieval against classic vector search — when is each right?
- Design a memory architecture for an agent that recalls facts across a 50-turn conversation without blowing up the context window.
- What is MCP and how do I let my RAG agent call an external tool or database safely?
- Give me an architecture blueprint for a production RAG service handling 10k questions/day on a single GPU.
- What security risks show up when agents can read private docs, and which guardrails matter?
- How do I evaluate whether my RAG system is good enough before launching?