/PubSec-Info-Assistant

Information Assistant, built with Azure OpenAI Service, Industry Accelerator

Primary LanguagePythonMIT LicenseMIT

Information Assistant Accelerator

Important

As of November 15, 2023, Azure Cognitive Search has been renamed to Azure AI Search. Azure Cognitive Services have also been renamed to Azure AI Services.

Table of Contents

Open in GitHub Codespaces

This industry accelerator showcases integration between Azure and OpenAI's large language models. It leverages Azure AI Search for data retrieval and ChatGPT-style Q&A interactions. Using the Retrieval Augmented Generation (RAG) design pattern with Azure OpenAI's GPT models, it provides a natural language interaction to discover relevant responses to user queries. Azure AI Search simplifies data ingestion, transformation, indexing, and multilingual translation.

The accelerator adapts prompts based on the model type for enhanced performance. Users can customize settings like temperature and persona for personalized AI interactions. It offers features like explainable thought processes, referenceable citations, and direct content for verification.

Please see this video for use cases that may be achievable with this accelerator.

Response Generation Approaches

Work(Grounded)

It utilizes a retrieval-augmented generation (RAG) pattern to generate responses grounded in specific data sourced from your own dataset. By combining retrieval of relevant information with generative capabilities, It can produce responses that are not only contextually relevant but also grounded in verified data. The RAG pipeline accesses your dataset to retrieve relevant information before generating responses, ensuring accuracy and reliability. Additionally, each response includes a citation to the document chunk from which the answer is derived, providing transparency and allowing users to verify the source. This approach is particularly advantageous in domains where precision and factuality are paramount. Users can trust that the responses generated are based on reliable data sources, enhancing the credibility and usefulness of the application. Specific information on our Grounded (RAG) can be found in RAG

Ungrounded

It leverages the capabilities of a large language model (LLM) to generate responses in an ungrounded manner, without relying on external data sources or retrieval-augmented generation techniques. The LLM has been trained on a vast corpus of text data, enabling it to generate coherent and contextually relevant responses solely based on the input provided. This approach allows for open-ended and creative generation, making it suitable for tasks such as ideation, brainstorming, and exploring hypothetical scenarios. It's important to note that the generated responses are not grounded in specific factual data and should be evaluated critically, especially in domains where accuracy and verifiability are paramount.

Work and Web

It offers 3 response options: one generated through our retrieval-augmented generation (RAG) pipeline, and the other grounded in content directly from the web. When users opt for the RAG response, they receive a grounded answer sourced from your data, complete with citations to document chunks for transparency and verification. Conversely, selecting the web response provides access to a broader range of sources, potentially offering more diverse perspectives. Each web response is grounded in content from the web accompanied by citations of web links, allowing users to explore the original sources for further context and validation. Upon request, It can also generate a final response that compares and contrasts both responses. This comparative analysis allows users to make informed decisions based on the reliability, relevance, and context of the information provided. Specific information about our Grounded and Web can be found in Web

Assistants

It generates response by using LLM as a reasoning engine. The key strength lies in agent's ability to autonomously reason about tasks, decompose them into steps, and determine the appropriate tools and data sources to leverage, all without the need for predefined task definitions or rigid workflows. This approach allows for a dynamic and adaptive response generation process without predefining set of tasks. It harnesses the capabilities of LLM to understand natural language queries and generate responses tailored to specific tasks. These Agents are being released in preview mode as we continue to evaluate and mitigate the potential risks associated with autonomous reasoning, such as misuse of external tools, lack of transparency, biased outputs, privacy concerns, and remote code execution vulnerabilities. With future releases, we plan to work to enhance the safety and robustness of these autonomous reasoning capabilities. Specific information on our preview agents can be found in Assistants.

Features

The IA Accelerator contains several features, many of which have their own documentation.

  • Examples of custom Retrieval Augmented Generation (RAG), Prompt Engineering, and Document Pre-Processing
  • Azure AI Search Integration to include text search of both text documents and images
  • Customization and Personalization to enable enhanced AI interaction
  • Preview into autonomous agents

For a detailed review see our Features page.

Process Flow for Work(Grounded), Ungrounded, and Work and Web

Process Flow for Chat

Process Flow for Assistants

Process Flow for Assistants

Azure account requirements

IMPORTANT: In order to deploy and run this example, you'll need:

  • Azure account. If you're new to Azure, get an Azure account for free and you'll get some free Azure credits to get started.
  • Azure subscription with access enabled for the Azure OpenAI service. You can request access with this form.
    • Access to one of the following Azure OpenAI models:

      Model Name Supported Versions
      gpt-35-turbo current version
      gpt-35-turbo-16k current version
      gpt-4 current version
      gpt-4-32k current version

      Important: Gpt-35-turbo-16k (0613) is recommended. GPT 4 models may achieve better results from the IA Accelerator.

    • (Optional) Access to the following Azure OpenAI model for embeddings. Some open source embedding models may perform better for your specific data or use case. For the use case and data Information Assistant was tested for we recommend using the following Azure OpenAI embedding model.

      Model Name Supported Versions
      text-embedding-ada-002 current version
  • Azure account permissions:
    • Your Azure account must have Microsoft.Authorization/roleAssignments/write permissions, such as Role Based Access Control Administrator, User Access Administrator, or Owner on the subscription.
    • Your Azure account also needs Microsoft.Resources/deployments/write permissions on the subscription level.
    • Your Azure account also needs microsoft.directory/applications/create and microsoft.directory/servicePrincipals/create, such as Application Administrator Entra built-in role.
  • To have accepted the Azure AI Services Responsible AI Notice for your subscription. If you have not manually accepted this notice please follow our guide at Accepting Azure AI Service Responsible AI Notice.
  • (Optional) Have Visual Studio Code installed on your development machine. If your Azure tenant and subscription have conditional access policies or device policies required, you may need to open your GitHub Codespaces in VS Code to satisfy the required polices.

Deployment

Please follow the instructions in the deployment guide to install the IA Accelerator in your Azure subscription.

Once completed, follow the instructions for using IA Accelerator for the first time.

You may choose to view the deployment and usage click-through guides to see the steps in action. These videos may be useful to help clarify specific steps or actions in the instructions.

Responsible AI

The Information Assistant (IA) Accelerator and Microsoft are committed to the advancement of AI driven by ethical principles that put people first.

Transparency Note

Read our Transparency Note

Find out more with Microsoft's Responsible AI resources

Content Safety

Content safety is provided through Azure OpenAI service. The Azure OpenAI Service includes a content filtering system that runs alongside the core AI models. This system uses an ensemble of classification models to detect four categories of potentially harmful content (violence, hate, sexual, and self-harm) at four severity levels (safe, low, medium, high).These 4 categories may not be sufficient for all use cases, especially for minors. Please read our Transaparncy Note

By default, the content filters are set to filter out prompts and completions that are detected as medium or high severity for those four harm categories. Content labeled as low or safe severity is not filtered.

There are optional binary classifiers/filters that can detect jailbreak risk (trying to bypass filters) as well as existing text or code pulled from public repositories. These are turned off by default, but some scenarios may require enabling the public content detection models to retain coverage under the customer copyright commitment.

The filtering configuration can be customized at the resource level, allowing customers to adjust the severity thresholds for filtering each harm category separately for prompts and completions.

This provides controls for Azure customers to tailor the content filtering behavior to their needs while aiming to prevent potentially harmful generated content and any copyright violations from public content.

Instructions on how to confiure content filters via Azure OpenAI Studio can be found here https://learn.microsoft.com/en-us/azure/ai-services/openai/how-to/content-filters#configuring-content-filters-via-azure-openai-studio-preview

Data Collection Notice

The software may collect information about you and your use of the software and send it to Microsoft. Microsoft may use this information to provide services and improve our products and services. You may turn off the telemetry as described in the repository. There are also some features in the software that may enable you and Microsoft to collect data from users of your applications. If you use these features, you must comply with applicable law, including providing appropriate notices to users of your applications together with a copy of Microsoft’s privacy statement. Our privacy statement is located at https://go.microsoft.com/fwlink/?LinkID=824704. You can learn more about data collection and use in the help documentation and our privacy statement. Your use of the software operates as your consent to these practices.

About Data Collection

Data collection by the software in this repository is used by Microsoft solely to help justify the efforts of the teams who build and maintain this accelerator for our customers. It is your choice to leave this enabled, or to disable data collection.

Data collection is implemented by the presence of a tracking GUID in the environment variables at deployment time. The GUID is associated with each Azure resource deployed by the installation scripts. This GUID is used by Microsoft to track the Azure consumption this open source solution generates.

How to Disable Data Collection

To disable data collection, follow the instructions in the Configure ENV files section for ENABLE_CUSTOMER_USAGE_ATTRIBUTION variable before deploying.

Resources

Navigating the Source Code

This project has the following structure:

File/Folder Description
.devcontainer/ Dockerfile, devcontainer configuration, and supporting script to enable both GitHub Codespaces and local DevContainers.
app/backend/ The middleware part of the IA website that contains the prompt engineering and provides an API layer for the client code to pass through when communicating with the various Azure services. This code is python based and hosted as a Flask app.
app/enrichment/ The text-based file enrichment process that handles language translation, embedding the text chunks, and inserting text chunks into the Azure AI Search hybrid index. This code is python based and is hosted as a Flask app that subscribes to an Azure Storage Queue.
app/frontend/ The User Experience layer of the IA website. This code is Typescript based and hosted as a Vite app and compiled using npm.
azure_search/ The configuration of the Azure Search hybrid index that is applied in the deployment scripts.
docs/adoption_workshop/ PPT files that match what is covered in the Adoption Workshop videos in Discussions.
docs/deployment/ Detailed documentation on how to deploy and start using Information Assistant.
docs/features/ Detailed documentation of specific features and development level configuration for Information Assistant.
docs/ Other supporting documentation that is primarily linked to from the other markdown files.
functions/ The pipeline of Azure Functions that handle the document extraction and chunking as well as the custom CosmosDB logging.
infra/ The Terraform scripts that deploy the entire IA Accelerator. The overall accelerator is orchestrated via the main.tf file but most of the resource deployments are modularized under the core folder.
pipelines/ Azure DevOps pipelines that can be used to enable CI/CD deployments of the accelerator.
scripts/environments/ Deployment configuration files. This is where all external configuration values will be set.
scripts/ Supporting scripts that perform the various deployment tasks such as infrastructure deployment, Azure WebApp and Function deployments, building of the webapp and functions source code, etc. These scripts align to the available commands in the Makefile.
tests/ Functional Test scripts that are used to validate a deployed Information Assistant's document processing pipelines are working as expected.
Makefile Deployment command definitions and configurations. You can use make help to get more details on available commands.
README.md Starting point for this repo. It covers overviews of the Accelerator, Responsible AI, Environment, Deployment, and Usage of the Accelerator.

References

Trademarks

This project may contain trademarks or logos for projects, products, or services. Authorized use of Microsoft trademarks or logos is subject to and must follow Microsoft’s Trademark & Brand Guidelines. Use of Microsoft trademarks or logos in modified versions of this project must not cause confusion or imply Microsoft sponsorship. Any use of third-party trademarks or logos are subject to those third-party’s policies.

Code of Conduct

This project has adopted the Microsoft Open Source Code of Conduct. For more information see the Code of Conduct FAQ or contact opencode@microsoft.com with any additional questions or comments.

Reporting Security Issues

For security concerns, please see Security Guidelines