This repo demonstrates how to query AzDO content using Langchain integration with Azure OpenAI.
Specifically, it defines custom document loaders for user stories and pull requests to facilitate Q&A like:
What is the title and owner of the most recently completed user story?
Summarize Alice's comments on Bob's PR for work item 234.
Note the techniques demonstrated in this repo are generalized and should be refined for specific use cases.
-
Access to an AzDO project and personal access token
-
An Azure OpenAI Service instance
-
A deployed embeddings model
-
A deployed GPT model (3.5 or 4)
-
Visual Studio Code and Docker Desktop for devcontainer support
-
Clone this repo and open VS Code in the repo root folder. When prompted, re-open as a devcontainer
-
In main.ipynb add your Azure OpenAI endpoint and key, as well as the names of your embeddings and GPT model deployments
-
Also in main.ipynb add your AzDO URI, PAT, and project name
-
In the last notebook cell, update the query as warranted for your use case
-
Run all cells and observe the results
-
Iterate on the query to coax relevant and accurate responses out of Azure OpenAI