Jagdish, thank you so much! 👍All other videos are bla-bla, you can do this, you can do that. With your two videos you have shown a simple and concise step by step procedure how to set up, and most importantly, how to include this within our apps. You have saved me many days of investigative work.😀Yes, langchain sample would be great, but you have already given us plenty.
What we'd like to implement is chat with different sources (different Sharepoints, intranet, public website, but also allow internet browsing for more general questions instead of restricting the answer to data in Azure Cognitive Search) and keeping history of course. We are using Power Virtual Agents, Power Automate and Azure OpenAI for the basic flow (storing chat history in Cosmos DB or maybe Dataverse, haven't decided yet). But for processing all those sources through Azure Cognitive Search, do you recommend bringing them all to a single Azure Blob Storage or something else? And how do you allow for flexibility so it doesn't restrict the responses to only those sources (for example, if I ask which is the largest river in the world, I expect it to retrieve the data from the internet and not those sources)? Also, would Microsoft Bot Framework would be better suited for what we are trying to achieve?
1. RAG on multi sources from Azure openai is still not supported for cognitive search. Means multi index in 1 api call is not supported yet. May be in the future because if you look at the API call request body, the sources is an array. 2. So, yes all needs to be in a single index for now for 1 source. Even if you keep your files in separate blob container, you can still index them in 1 index through coding. 3. For searching internet, change the value of inScope to false instead of true. This will look at your documents as well as internet (Cutoff to Sept 2021). 4. If you want to customize your implementation with some extra business logic, then bot framework sdk is preferrred. If only calling the api to azure openai with index, then PVA would be easy to go with.
I get the same issue. When i hit the api which gets stuck on fetching. It sometimes takes alot of time for the model to reply back. I thought i was the only one getting it. Glad that I wasn’t the only one.
I have been attempting to implement "Chat with Azure OpenAl models using your own data" with a large dataset stored inside different folders in a blob storage. Each time I attempt to execute the indexer, it encounters failure due to the large size of the PDF files. Although I've utilized the provided Data Preparation script, it necessitates downloading the files to the local machine and re-uploading them. Is there any workaround to address this issue?
Scripts are recommended for large docs. Yes, it downloads to local machine and uploads to index after reading using form recognizer. You can modify the script to delete the files after indexing. I did same for my API developed using those scripts.
I would love to follow along with your echobot implementation... I was looking for the repo online but couldn't find code that matches yours. Could you share your echobot repo? Thank you for your informative videos!
Jagadish, Lets say If I have some SDK & sample code snippets on how to use my SDK ( kind of help/manual pages ) , i want to build a custom pilot so that my SDK users can ask sample codes to copilot and get my sample code snippets. Most of the content i see is to train some text data ? but can you help me to understand on how to train my code snippets ?
Also, you can just create a function which removes the test data name while you implement it. Easily doable in javascript. Also can be implemented in your website. I created a function which just removes everything in [] brackets as none of the data has [] brackets.
You hit this api which you copied from azure ai studio can we also deploy that into app service and consume that api if yes then will that api also be called same way
Is it good to deploy it into app service or directly call like how you did what are the advantages of deploying and directly calling and also one thing the way you called api here whether it will take history conversations or not?
Yes, same api will work if you deploy in app service. Yes, this API is capable of managing history. Your application must manage that and pass it along the json body.
@@Dewiride thanks for the reply i am not deploying and using how you used . can you pls pls guide me how to pass form body with history in it in dynamic manner not harcoded it will be very helpful for me { "dataSources": [ { "type": "AzureCognitiveSearch", "parameters": { "endpoint": "", "key": "", "indexName": "finaldemo", "semanticConfiguration": "", "queryType": "simple", "fieldsMapping": null, "inScope": true, "roleInformation": "You are an AI assistant that helps people find information." } } ], "messages": [ { "role": "user", "content": "What is the emergency landline number for medical in germany?" } ], "deployment": "gpt-35-turbo", "temperature": 0, "top_p": 1, "max_tokens": 800, "stop": null, "stream": false } This is normal i want with history
@@mathematicmadeasy for history in dynamic way, you need to manage that in your application. Here is one example of managing history using Storage Account - ruclips.net/video/0joaIgHjbEk/видео.html
Jagdish, thank you so much! 👍All other videos are bla-bla, you can do this, you can do that. With your two videos you have shown a simple and concise step by step procedure how to set up, and most importantly, how to include this within our apps. You have saved me many days of investigative work.😀Yes, langchain sample would be great, but you have already given us plenty.
Thank you so much 😀
11:20 you can directly click on beautify button in postman to format json payload
Thanks for the suggestion. Still learning some editor features.
This is so cool! Thank you for making a video on this 😊
You're so welcome!
thanks a ton for the training
Your welcome.
What we'd like to implement is chat with different sources (different Sharepoints, intranet, public website, but also allow internet browsing for more general questions instead of restricting the answer to data in Azure Cognitive Search) and keeping history of course. We are using Power Virtual Agents, Power Automate and Azure OpenAI for the basic flow (storing chat history in Cosmos DB or maybe Dataverse, haven't decided yet). But for processing all those sources through Azure Cognitive Search, do you recommend bringing them all to a single Azure Blob Storage or something else? And how do you allow for flexibility so it doesn't restrict the responses to only those sources (for example, if I ask which is the largest river in the world, I expect it to retrieve the data from the internet and not those sources)? Also, would Microsoft Bot Framework would be better suited for what we are trying to achieve?
1. RAG on multi sources from Azure openai is still not supported for cognitive search. Means multi index in 1 api call is not supported yet. May be in the future because if you look at the API call request body, the sources is an array.
2. So, yes all needs to be in a single index for now for 1 source. Even if you keep your files in separate blob container, you can still index them in 1 index through coding.
3. For searching internet, change the value of inScope to false instead of true. This will look at your documents as well as internet (Cutoff to Sept 2021).
4. If you want to customize your implementation with some extra business logic, then bot framework sdk is preferrred. If only calling the api to azure openai with index, then PVA would be easy to go with.
I get the same issue. When i hit the api which gets stuck on fetching. It sometimes takes alot of time for the model to reply back. I thought i was the only one getting it. Glad that I wasn’t the only one.
Is it possible to upload multiple documents and query through the same ?
Yes
I have been attempting to implement "Chat with Azure OpenAl models using your own data" with a large dataset stored inside different folders in a blob storage. Each time I attempt to execute the indexer, it encounters failure due to the large size of the PDF files. Although I've utilized the provided Data Preparation script, it necessitates downloading the files to the local machine and re-uploading them. Is there any workaround to address this issue?
Scripts are recommended for large docs. Yes, it downloads to local machine and uploads to index after reading using form recognizer.
You can modify the script to delete the files after indexing. I did same for my API developed using those scripts.
I would love to follow along with your echobot implementation... I was looking for the repo online but couldn't find code that matches yours. Could you share your echobot repo? Thank you for your informative videos!
Will be available soon
Here's the source code used - shop.dewiride.com/product/calling-api-for-train-your-own-enterprise-data-with-azure-openai-service-using-c/
Can we do this without Azure openai ? Can we use public openai key to do this?
Yes, we can using Langchain and Vector DBs or other Indexing DBs.
Hie. Whats makes it say it does not find information regarding a specific question even tho its there clearly in the provided data ?
Try changing the prompt or use semantic search.
This video uses semantic search - ruclips.net/video/mtXxtSZsIpI/видео.html
Thank you . Let me check it out now
Thankyou for the video. Love your videos. What about langchain implementations..Would love to see that too.
Noted
Jagadish, Lets say If I have some SDK & sample code snippets on how to use my SDK ( kind of help/manual pages ) , i want to build a custom pilot so that my SDK users can ask sample codes to copilot and get my sample code snippets. Most of the content i see is to train some text data ? but can you help me to understand on how to train my code snippets ?
The codex models was available earlier and now discontinued. I believe you can still use text based model to train your code snippets.
Hey @Dewiride how can I restrict my search to uploaded documents.btw your solution works like a charm
Use the filterable feature in Azure Cognitive Search Fields and use queries to filter/restrict data
Also, you can just create a function which removes the test data name while you implement it. Easily doable in javascript. Also can be implemented in your website. I created a function which just removes everything in [] brackets as none of the data has [] brackets.
You hit this api which you copied from azure ai studio can we also deploy that into app service and consume that api if yes then will that api also be called same way
Is it good to deploy it into app service or directly call like how you did what are the advantages of deploying and directly calling and also one thing the way you called api here whether it will take history conversations or not?
Yes, same api will work if you deploy in app service. Yes, this API is capable of managing history. Your application must manage that and pass it along the json body.
@@Dewiride thanks for the reply i am not deploying and using how you used .
can you pls pls guide me how to pass form body with history in it in dynamic manner not harcoded it will be very helpful for me
{
"dataSources": [
{
"type": "AzureCognitiveSearch",
"parameters": {
"endpoint": "",
"key": "",
"indexName": "finaldemo",
"semanticConfiguration": "",
"queryType": "simple",
"fieldsMapping": null,
"inScope": true,
"roleInformation": "You are an AI assistant that helps people find information."
}
}
],
"messages": [
{
"role": "user",
"content": "What is the emergency landline number for medical in germany?"
}
],
"deployment": "gpt-35-turbo",
"temperature": 0,
"top_p": 1,
"max_tokens": 800,
"stop": null,
"stream": false
}
This is normal i want with history
@@mathematicmadeasy for history in dynamic way, you need to manage that in your application. Here is one example of managing history using Storage Account - ruclips.net/video/0joaIgHjbEk/видео.html