Hi. I’m an AI engineer of an emerging retailer. We’re continuously pushing the boundaries of our user search experience. We’ve got a massive inventory, hence a lot of data to be managed. This got me thinking about the untapped potential of neural search.

I’ve had my hands on OpenAI’s GPT and Deepset’s Haystack lately. Both tools are great in specific scenarios, but integrating them seamlessly at an enterprise scale is challenging, especially when we’re talking about real-time user interactions. The primary challenge remains in managing multimodal data efficiently without sacrificing speed.

To add context, my goal in leveraging something like GPT for e-commerce is to create a more intuitive, conversational, and responsive search function. Imagine a user typing in a vague description or query, and the system providing product suggestions like a seasoned salesperson would. Given the vast product range, the neural search could bridge the gap between user intent and the most relevant product offerings.

If anyone has experience with this I’d like to hear your thoughts, and if you have any other tool recommendations for this pls do share. I’d be grateful for any help

  • Seankala@alien.topB
    link
    fedilink
    English
    arrow-up
    1
    ·
    1 year ago

    Is there any reason why you won’t just use a CLIP-based model and why you’re trying to use OpenAI’s GPT?

    I’m also in charge of a text-image (text-image, not multimodal in my case) model that my company’s trying to create a search product with. There have been talks about using “ChatGPT” from higher-ups but I just don’t see the reason why we’d have to do this. I figured that a simple NER model or something would work just as well, I mean how many people do online shopping while expecting textual responses from the website.