Using ChatGPT with Product Data

I work for an ecom company and have been looking into different OpenAI solutions, but I was hoping for some advice about how to achieve the following:

We can export a feed of all our products (JSON, including description, tags, spec, media urls etc.). I would like to know what the most economical way of using this product feed with ChatGPT would be - so that we can describe a type of person / occasion and then ChatGPT would give the top products it recommends from the selection of products we have. This could just be simply returning e.g. the product IDs of 10 products

The product feed is a ~1mb JSON document of about 1000 products- although could be a JSONL or something else. It would probably need to be updated weekly.

I’ve got this to work by attaching the file but I think the cost was looking too high as it was perhaps sending the full document with each request? I’ve looked at assistants and embeddings but I’m still not sure if they’re right for my use-case.

Essentially, I’d like use a node app to make and receive requests.

An example of the experience would be like “a birthday gift for dad” and ChatGPT would return the product IDs of the most relevant gifts that we carry like a bottle of whisky, a flask, a fishing rod, etc.

Any help for how best to achieve this would be appreciated, thanks!

Hi @michael84 , based on your use-case, I would suggest using an Assistant with file search capabilities. With good prompting and testing, it should help you with good output. gtp-4o-mini is inexpensive for such use cases. In case the assistant hallucinates, split the file with 100 products per file and upload in the vector store for assistant to use.

If assistant responses are not upto the mark, try fine-tuning. Though I think the assistant would be good enough.

1 Like

Thank you @bhagyesh that sounds promising, I’ll look into this further.