Daikin Asistent

Embedding Model:

OpenAI embedding model for vector search

Chunk Size:

Number of characters per document chunk (1000-2000 recommended)

Chunk Overlap:

Characters of overlap between chunks (200-400 recommended)

Number of Retrieval Chunks:

How many document chunks to retrieve for context (6-12 recommended)

Response Model:

OpenAI model for generating responses (gpt-4.1, gpt-3.5-turbo, etc.)

Max Response Tokens:

Maximum length of generated responses (400-1000 recommended)

Temperature:

Randomness in responses (0 = deterministic, 1 = creative)

System Prompt (Grounding):

Main instructions for AI. Use {context} as placeholder for retrieved documents.

Behavioral Prompt:

Additional behavioral guidelines and formatting instructions.

Bok! Ja sam Danko, tvoj AI asistent!