Answer Requests for Proposals with our own data

Question

Hello,

Here is my challenge. We are answering a lot of Requests for Proposals for potential clients. These RFPs contain scenarios and questions related to the project. I want to use previous Answers to RFPs we did to automate the process of answering the different questions. I figured I will need to upload between 15 to 20 documents (about 25 pages each) to build Azure OpenAI knowledge.

I am not sure how to train Azure AI on the documents I uploaded.
So far I created:
Azure Storage Account
Azure Cognitive Search
Azure Blob where my documents are uploaded
An Index
Azure Open AI service

I used the Azure OpenAI Chat and deployed a web app but the bot only returns textual part of the document. Is there a way to improve the model? Can it learn from previous Answers to RFPs.

Sorry I'm all over the place !

Thanks for any help you can provide.

Max

Answer 1

Maxime Viens Greetings & Welcome to Microsoft Q&A forum!

I used the Azure OpenAI Chat and deployed a web app but the bot only returns textual part of the document.

Are you following any documentations? Azure OpenAI Service lets you tailor the models to your personal datasets by using a process known as fine-tuning. This customization step lets you get more out of the service by providing:

• Higher quality results than what you can get just from prompt engineering
• The ability to train on more examples than can fit into a model's max request context limit.
• Lower-latency requests, particularly when using smaller models.

A fine-tuned model improves on the few-shot learning approach by training the model's weights on your own data. A customized model lets you achieve better results on a wider number of tasks without needing to provide examples in your prompt. The result is less text sent and fewer tokens processed on every API call, potentially saving cost and improving request latency.

Azure OpenAI on your data provides several search options you can use when you add your data source, leveraging the following types of search.

Keyword search

Semantic search

Vector search using Ada embedding models, available in select regions.

Yes, you can use Azure OpenAI to train a model on the documents you uploaded and use it to automate the process of answering different questions. To train the model, you can use the Azure OpenAI fine-tuning workflow.

I am not sure how to train Azure AI on the documents I uploaded. Is there a way to improve the model?

See Customize a model with fine-tuning (preview) for suggestions on how to train the models.

Regarding your question about improving the model, you can improve the model by fine-tuning it on more data. You can also try different base models to see which one works best for your use case. See Azure OpenAI on your data (preview) for more details.

Can it learn from previous Answers to RFPs.

To answer your question, yes by fine-tuning the model on the previous answers to RFPs, you can teach the model to recognize patterns and generate answers to new questions based on the patterns it has learned.

I hope this helps. Do let me know if you have any further queries.

---

If the response helped, please do click Accept Answer and Yes for was this answer helpful.

Doing so would help other community members with similar issue identify the solution. I highly appreciate your contribution to the community.