want to do q&A with azure open ai search with index and indexer on JSON data alone only.

Question

want to do q&A with azure open ai search with index and indexer on JSON data alone only.

Polisetty, Balaji 0

I want to do azure open ai search on json data alone. i have huge data on json data only.

i created index and indexer.

what is best way to ask question on json data with GPT 4 model.

 "json": {
        "cellbattery_Level": "100",
        "cellBattery_Cap": "3221",
      
}

if i ask question whats is the battery level i am not getting answer if i ask for cellbattery_Level i am able to get answer.

what should i do to get answers if i ask for cellbattery_Level .

one more question -

is there any possiblity to have prediction questions based on the previous answer.

Like if the question is how is my code looking ? prompts like its good / i have review comments /its not good/ its bad. do we have any option to do like that ?

Polisetty, Balaji 0 Reputation points

2023-12-19T17:28:39.7933333+00:00

thank you predictive questions in the sense - i am using azure open ai rest api to ask my questions and i am getting answers for now. is there any possibility to get predictive small questions on the last answer like I am fine / I am good / etc..,
Polisetty, Balaji 0 Reputation points

2023-12-19T18:00:15.98+00:00

while creating datasources all the json in the blob should be of same format like same set of keys ?

lets say i have couple of files having one set of keys and other files having different set of key .. like that we different group do we need mention all in the index creation for proper search should happen
SnehaAgrawal-MSFT 22,706 Reputation points Moderator

2023-12-21T02:46:16.0233333+00:00

@Polisetty, Balaji If you prompt correctly and maybe even persists chat history you can have a contextual conversations.

JSON blob format, you should have the same format of json blobs then connect that as a data source in Azure AI Search, if you have different formats you can create different data sources and indexers to field map to your search index in whatever way you desire.

If possible, a best practice is to have a json schema and search index that are identical and then you can simply map 1:1 from data source to index.
SnehaAgrawal-MSFT 22,706 Reputation points Moderator

2023-12-22T05:39:50.96+00:00

@Polisetty, Balaji Just checking if you have chance to see recent response.

Please let us know if further query or issue remains.

Please accept as "Yes" if the answer provided is useful, so that you can help others in the community looking for remediation for similar issues.
SnehaAgrawal-MSFT 22,706 Reputation points Moderator

2023-12-26T06:08:05.55+00:00

@Polisetty, Balaji A gentle follow-up, let us know if further query or issue remains.

Please accept as "Yes" if the answer provided is useful, so that you can help others in the community looking for remediation for similar issues.
Polisetty, Balaji 0 Reputation points

2023-12-26T14:28:04.0833333+00:00

Hi Sneha - I am able to work out with vector indexing and all i want is context based questions from rest end point. how i can use persist chat history with rest end point.

one more question - as i already asked is there any easy way of having predicative questions based on previous answers. how can i implement that in my react web app with azure ai rest end points.

thanks
SnehaAgrawal-MSFT 22,706 Reputation points Moderator

2023-12-28T07:28:34.57+00:00

@Polisetty, Balaji Thanks for response, Please reach out to us by sending an email with subject line “Attn:Sneha |thread title” to AzCommunity[at]microsoft[dot]com referencing this thread, for closer look on this.
Polisetty, Balaji 0 Reputation points

2024-01-05T03:51:47.13+00:00

i tried sending a mail to AzCommunity@microsoft.com but mail failed send to receipent.
SnehaAgrawal-MSFT 22,706 Reputation points Moderator

2024-01-05T10:30:27.5533333+00:00

@Polisetty, Balaji Thanks received! Could you please share asked details to help you better.

1 answer

Your answer

Polisetty, Balaji 0 Reputation points

2023-12-19T17:28:39.7933333+00:00

thank you predictive questions in the sense - i am using azure open ai rest api to ask my questions and i am getting answers for now. is there any possibility to get predictive small questions on the last answer like I am fine / I am good / etc..,
Polisetty, Balaji 0 Reputation points

2023-12-19T18:00:15.98+00:00

while creating datasources all the json in the blob should be of same format like same set of keys ?

lets say i have couple of files having one set of keys and other files having different set of key .. like that we different group do we need mention all in the index creation for proper search should happen
SnehaAgrawal-MSFT 22,706 Reputation points Moderator

2023-12-21T02:46:16.0233333+00:00

@Polisetty, Balaji If you prompt correctly and maybe even persists chat history you can have a contextual conversations.

JSON blob format, you should have the same format of json blobs then connect that as a data source in Azure AI Search, if you have different formats you can create different data sources and indexers to field map to your search index in whatever way you desire.

If possible, a best practice is to have a json schema and search index that are identical and then you can simply map 1:1 from data source to index.
SnehaAgrawal-MSFT 22,706 Reputation points Moderator

2023-12-22T05:39:50.96+00:00

@Polisetty, Balaji Just checking if you have chance to see recent response.

Please let us know if further query or issue remains.

Please accept as "Yes" if the answer provided is useful, so that you can help others in the community looking for remediation for similar issues.
SnehaAgrawal-MSFT 22,706 Reputation points Moderator

2023-12-26T06:08:05.55+00:00

@Polisetty, Balaji A gentle follow-up, let us know if further query or issue remains.

Please accept as "Yes" if the answer provided is useful, so that you can help others in the community looking for remediation for similar issues.
Polisetty, Balaji 0 Reputation points

2023-12-26T14:28:04.0833333+00:00

Hi Sneha - I am able to work out with vector indexing and all i want is context based questions from rest end point. how i can use persist chat history with rest end point.

one more question - as i already asked is there any easy way of having predicative questions based on previous answers. how can i implement that in my react web app with azure ai rest end points.

thanks
SnehaAgrawal-MSFT 22,706 Reputation points Moderator

2023-12-28T07:28:34.57+00:00

@Polisetty, Balaji Thanks for response, Please reach out to us by sending an email with subject line “Attn:Sneha |thread title” to AzCommunity[at]microsoft[dot]com referencing this thread, for closer look on this.
Polisetty, Balaji 0 Reputation points

2024-01-05T03:51:47.13+00:00

i tried sending a mail to AzCommunity@microsoft.com but mail failed send to receipent.
SnehaAgrawal-MSFT 22,706 Reputation points Moderator

2024-01-05T10:30:27.5533333+00:00

@Polisetty, Balaji Thanks received! Could you please share asked details to help you better.

Answer 1

@Polisetty, Balaji Thanks for reaching here! To enable natural language queries on JSON data using Azure AI Search with a GPT-4 model, you would typically need to configure the search service to understand and process natural language questions.

You can index JSON blobs similar to the sample and leverage the indexer to ingest these JSON blobs into a search index. Alternatively, you can use the Push API to upsert your json into a search index. Here are some high level steps-

Index Configuration:
Ensure that the fields in your index are set up with the correct attributes (searchable, filterable, sortable, etc.) depending on how you want to query them
Indexer Configuration (if using 'pull' approach):
When setting up the indexer, map your source data fields to the appropriate index fields.
- You can use field mappings to transform property names in the source data to more user-friendly field names in the index that could be better suited for natural language queries.
- Also encourage you to take advantage of creating vector representations of your data so you can benefit from vector and/or hybrid search.
- You can take advantage of the "Import and Vectorize" feature in the Portal and Azure AI Search will handle this process for you.
- Integrate with Azure OpenAI service:
- In order to ground your data to a model like GPT-4, they are multiple methods of doing this. You can use tools like LangChain, Llamaindex, and Semantic Kernel to have an orchestration to demonstrate the RAG process using your Azure AI search index and Azure OpenAI service with GPT-4 - IMHO, this is more of a code-first approach. Example repo: azure-search-vector-samples/demo-python/code/azure-search-vector-python-llamaindex-sample.ipynb at main · Azure/azure-search-vector-samples (github.com)
- Alternatively, you can use Azure AI On Your Data and add your search index and begin chatting with it in natural language (Using your data with Azure OpenAI Service - Azure OpenAI | Microsoft Learn) - IMHO, this is more of a no-code approach
- Try testing your desired queries to see if you are getting the desired results back. If not, you may need to iterate on some advanced RAG strategies such as leveraging an LLM to write a query to Azure AI search that can ask for the cell battery given an ID. Re: Predictive Questions
For predictive queries like "how is my code looking?" you would need a model that understands the context of code quality, which might be beyond the scope of a typical search index.
You could potentially use the OpenAI Codex model, which is designed to understand and generate code, and integrate it similarly to how you might use GPT-4.

Hope this helps.

Share via

want to do q&A with azure open ai search with index and indexer on JSON data alone only.

1 answer

Your answer