Core Concepts
Understanding how Sela's AI works, from RAG to escalation.
RAG (Retrieval Augmented Generation)
Sela uses a technique called RAG to answer customer questions accurately.
- Ingestion: You upload PDFs, Docs, or URLs.
- Indexing: Sela breaks this content into small "chunks" and creates vector embeddings.
- Retrieval: When a user asks a question, we find the most relevant chunks.
- Generation: GPT-4 generates an answer based only on those chunks.
Escalation
AI isn't perfect. Sela has built-in triggers to hand off conversations to human agents.
Sentiment Analysis
If a customer uses angry language or seems frustrated, the AI instantly attempts to escalate.
Unknown Answer
If the answer isn't in your knowledge base, Sela will say "I don't know" and offer human help.