Connect ChatGPT to your Enterprise Data using Cognitive Search

[@EnseñanzasZenBudistas]

As someone else mentioned, having a comprehensive video tutorial on setting up this demo app would be incredibly beneficial. It would provide us with a deeper understanding of the concept and make the process much clearer. Thank you in advance for considering this request!

[@abelrugi5]

Exactly why do they beat about the bush

[@1242elena]

YES!

[@Laughbankrip]

I recently presented this at an AI hackathon within my company and it was well-received.

[@magicofjafo]

Oh man, I have a project that includes searching a ton of small documents. And this is the exact information I need.

[@CaioCosta-sf7hn]

It would be very helpful if you could make a video tutorial, step by step, setting up this infrastructure in Azure and using your own PDF to customize the model.

[@RaynardL]

Good to see you doing great things with your knowledge Pratheek!!

[@PD_CloudTech]

Thank you so much!

[@louisamakye8561]

great video. where can i find the notebook in the repo

[@Idlepug]

Point of note for enterprise use, be sure to be clear on the Microsoft ToS for Azure OpenAI service and how Microsoft employees can potentially end up reviewing your company data processed in the service if it is flagged by their abuse monitoring service (potential for false positives). Microsoft do make it pretty clear on this in their Azure documentation. There is an exemption form that can be submitted to Microsoft which, if they approve it, will exclude the subscription's OpenAi service from abuse monitoring.

[@Marco_Antonio_CV]

Excellent video, hope we have more like this soon

[@SQLKC]

I would like to see a demo using some public database like large city traffic data into ChatGPT. Like some data set with millions of rows, hundreds of tables

[@matheahildre4278]

Has anyone tried something like this with hundreds or thousands of documents? I wonder how well the querying for relevant documents work, especially within a specific domain. Does it manage to find the most relevant documents?

[@davidchoi1655]

This really helps! Thanks.

[@z-ro]

Fantastic video!! Your explanation is so clear and thorough. I keep coming back to this video to get a better understanding of Azure Openai and Cognitive Services. Just one question, when you ran the repo locally, did you have to configure the AZURE_PRINCIPAL_ID in your .env file? I was able to get the repo running locally but the API requests are failing... and I'm wondering if its because I'm missing this environment variable that I don't have access to.

[@adithyaskolavi]

A really informative video helped out a lot

[@akashvarkala4040]

love your videos

[@TheLeads]

If I add a new document in the storage will the model identify it and include in the search or I had to again drop and retrain the model?

[@doncristobal33]

Amazing, keep it up!

[@AxelJimenezC]

I tried this architecture but it works terrible for low volume data. To bug it, you only have to change a few words so that the architecture responds as if there were no information in the knowledge database. And the architecture forces us to split the PDF into pages, so there are more errors when the information is contained in two pages (there is no context in the response when you start with some context and the first page and the pdf has two pages). the only way i solved these problems is to pass the information to gpt to make multiple versions of the document with different layout. Or in an easy way you can use Embeddings with llama-index and remove azure search engine which is not the best way to embed data to gpt.

[@nipunwalia6990]

hey there, I was experimenting around this and found out that it cannot analyse the context of data. For eg. I trained it on 10-k of amazon and asked which sector made the most profit, it was unable to answer but if I asked the information with a targeted question like profit earned in automation sector, It gave a accurate answer. Any idea how can I work it out

[@albertliu5760]

thanks the update

[@amirbadawi3548]

since there is no powershell on mac does that means that this code won't work on arm based chips?

[@JBuckk]

This looks interesting. Do you have Experience with langchain aswell?

[@PD_CloudTech]

I'll try to make a video on langchain next time :)

[@JBuckk]

@@PD_CloudTech all good i was just wondering ^^

[@vigneshnagaraj7137]

@@PD_CloudTech Please make a video on langchain.

[@aitou96]

Is it possible that chatgpt/ gpt-4 can give answers to users regarding KPI structure in a dashboard by giving access to the data sources (raw data) like SAP etc. ?

[@DatumRadixx]

yes

[@amansaxena4314]

Yes, this is the exact usecase we are working on!

[@jimmynguyen3386]

Hi! Great video. I’d like to incorporate Open AI service in my org and I’m looking to hire a consultant to get us up and running. I didn’t see any contact info on your channel but if you’re interested, I’d love to chat

[@77netwar]

I like the video and the contents, but I would very appreciate it if someone at Microsoft shows from start to finish how to make a Azure environment, install OpenAI, Cognitive Search and other things that are needed, how to periodically upload my company data from a file server or other location, and how to make a website/webinterface that my company or someone from the outside can access. No offense, but I have seen four Microsoft youtube videos now with the same content that showcase this demo, but nothing goes into depth. I do not understand this. At the moment everybody is trying to race towards the finish line to get such a service running (especially using open source models like MPT-7B and Llama derivatives and interfaces like oogabooga and pipelines like Langchain) and Microsoft claims that they have the pipeline already there, but there is no in depth instructional video. If you make it a course or something, people will be happy to pay, including me.

[@syogeeshwari4998]

kzbin.info/www/bejne/amKnhKqKjrd4hNU

[@VageeshPrasad]

Great video. Can you clarify how can we filter the contents of AI response based on specific users and specific documents. For example, user 1 is allowed to view Document 1 and, user 2 is allowed to view Document 3 and user 3 is allowed to view Document 2 and 3. In this situation, each user should be allowed to see responses coming from documents that they are allowed to see.

[@Pure_Science_and_Technology]

Seems expensive. All Microsoft….