Talk to document or website and have it talk back

We have been creating a tool where you can upload a document or website link and then talk to it. You can find it here:

Was looking at the use cases we can target with this tool. Was wondering what the community here think of this?

1 Like

From the automotive sector, I’ve been dying to find an easy way to extract the important/relevant stuff out of Vehicle Manufacturer repair methods. The issue is the amount of context the model would need around the repair to do this effectively.

Thought I’d throw it out there in-case anyone has ideas.

1 Like

Wow, this is a great use case. Can I have a sample document which I create a demo with. Would love to see how well it works with this.

1 Like

Do you have a public manual that I can try this on? Like a car manual or engine manual? Anything that is ok to be public?

I can create a POC?

1 Like

I have the same thought!

I have a database of repair PDFs. Retrieval wasn’t reliable enough (yet) to provide good information. I would seelect and attach the respective manual for vehicle and it was good, but prone to hallucinations.

Three times it told me to torque the calipers mounting bolts to something dangerous like 20ftlb. It was also expensive AF to run.

I know OpenAI said they’ll protect anyone who gets sued for copyright. I don’t think they’ll have my back if my GPT tells someone to barely torque vital parts :sweat_smile: . But, it’s improving fast. Davinci tried telling me to take the brake hose straight off the calipers (for no reason) and then convinced me that I didn’t need to bleed them afterwards!

I thought about distilling the manuals. Values and factual information as JSON, instructionals retained as unstructured text. It’s on the backburner

1 Like

Try this and let me know how it goes.

You’re okay to make the Science Behind Bumper Repairs doc public if you want, that ones freely available.

The issue I seem to find is a large amount of the information in methods is in images, not text. This poses an issue when you need to know which bolts in the image can’t be reused.

Ya, fidelity of response has been an issue. It is getting really better though.

With the recent Assistant API that was released last week, it has got really better now. We will try to create a demo.

We are trying to also give reference in the actual document when users talk to it so that they can check the document at that place when they have an inssue.

If you have a sample document that is public, let us know. We will be happy to create a demo.


Thank you for these documents. We will work with this. Will get back.

1 Like

Have made a demo where you can talk to this document, see it here:

Click on the microphone to start talking with it. The microphone needs to be on.

1 Like