Sportski pozdrav! Ja sam Deki, vaš AI asistent.

Uz mene, sve informacije o opremi za skijanje dostupne su vam odmah - jednostavno, pouzdano i bez čekanja.

⚙️ Configuration

Re-indexing in progress...

🔄 Indexing Settings (requires re-indexing)

Embedding Model:

OpenAI embedding model for vector search (we embed whole products; no chunking)

🔍 Retrieval Settings

Retrieval Chunks:

Number of chunks to retrieve (6-12 recommended)

Rerank Top N:

How many retrieved chunks/products to keep after reranking

Enforce must_have_terms (strict feature filtering) When enabled, required features (e.g. MIPS/BOA/vizir) are enforced deterministically on retrieved docs. Turn OFF to test behavior without strict must-have filtering.

🤖 LLM Settings

LLM Model:

OpenAI model (gpt-4o-mini, gpt-4o, gpt-3.5-turbo, etc.)

Filter Extraction Model:

OpenAI model used ONLY for filter extraction (tool calling).

Max Tokens:

Maximum response length (400-1000 recommended)

Temperature:

Creativity (0 = deterministic, 1 = creative)

💬 System Prompts

Prompts live in prompts/*.txt, not in the settings file. Edit the files, then press Reload prompts to apply them without restarting the server.

System Prompt (prompts/system_prompt.txt) Main instructions. Use {context} for retrieved documents.

Behavioral Prompt (prompts/behavioral_prompt.txt) Behavioral guidelines and formatting instructions.

🛒 Košarica

Košarica je prazna.
Dodajte proizvode klikom na "Dodaj" gumb pokraj preporuka.

Ukupno: 0,00 €