Inspiration -To learn about Gen AI ,cosmos DB and vector search and prepare a custom copilot

What it does - For phase 1 the copilot is trying to answer queries by the user related to a bike inventory whose data is loaded in cosmos db collection

How I built it - Basically we are integration azure open ai with azure cosmos DB, so initially cosmos DB with mongo db vcore resource is created ,also azure open ai service resource is created where in deployments we have added 'completions' and 'embeddings' model . Initially we are loading the product data in cosmos DB collection in form of json ,we are creating VectorSearchIndex(ivf) for the collection ,vector embeddings are generated through the embedding model based on the data attributes, vector search is performed on specified collection created by vectorizing

the query and searching the vector index for the most similar documents, the chat completions using the AzureOpenAICompletionClient and the response in formed from the vector search results. Apart from performing the action through code, I have performed BYOD option where in the data source I added the below: Data source:

Azure Cosmos DB for MongoDB vCore Index:

VectorSearchIndex Database name:

Products Advanced settings then I deployed the model as webapp from deploy to webapp in azure

Challenges I ran into -I ran into many challenges initially Because the steps provided in the exercise is really difficult to follow for someone who is new into this area ,none of the steps were explained clearly ,so I had to search multiple videos to understand the flow and complete my challenge.

Accomplishments that I'am proud of -able to complete my challenge in a week in spite of being very new in this area

What I have learned -gained Understanding in azure open AI services , understanding on cosmos DB, RAG pattern and how we can use cosmos DB in RAG pattern implementation

What's next for Submission for deliverables in Phase2 - Screenshots for phase 1 are attached here ,will share the app details and other deliverables for pahse2 shortly.

username-gosmiadri@gmail.com

Built With

Share this project:

Updates