Hi! I’m creating a workflow to compare proposals from suppliers we’ve requested quotes from.
The process works like this: the buyer enters the quantity and description of the products they requested:
50 8-gauge green cable
50 8-gauge red cable
50 8-gauge black cable
Using an embedding node, the data is vectorized and saved in Supabase. Then, when the client uploads a quote, another workflow performs OCR to extract the descriptions and quantities.
Each description is vectorized to compare it with the five most similar products the client added to the database, allowing for comparison by an LLM (Linear Material Manager).
Everything works quite well, the problem is that the suppliers provide very diverse quotes. For example, if we request 50 meters of three different types of cable, they group them into 150 meters of one type of cable.
So in the ddel llm chain prompt I added 3 possible outputs: Match, new_item, and exception. The problem is that it throws it into exception too many times when it shouldn’t. I don’t know if this is the best way or if you can think of a more logical way to avoid this.
Do you think good prompting will be enough? It’s for a client, and I don’t want to take any risks.
Thank you.