Secure AI-Driven Data Visualization & Reporting

Tech Stack

Azure OpenAI services, Azure Synaptic Analytics, Azure SQL Storage, Azure Data Factory, Microsoft 365, Python, Streamlit

Problem

The client needed a custom AI solution to let his employees create visuals using natural language and their own data, but he was worried about the data being leaked and used to train public AI algorithms.

Solution

1. We implemented a solution for our client which didn’t require the client to share their data with the AI models, create visually appealing graphs and share them to external users, stakeholders and internal employees hassle free. 
2. We used the Azure OpenAI services to access the GPT- 4 Model. We utilized the model to generate the summary reports for each visual and the python code required to generate the visuals.
3. We also used Azure SQL Storage, Data Factory and Synaptic Analytics to store and process client's data.
4. We created a custom web app to let the client visualize the graphs on any device as well as share them with external stakeholders and internal employees. We used Azure Web Services to host our Web App. 
5. The web app was deeply integrated with the Microsoft 365 ecosystem letting the employees directly share the generated graphs to the platform of their preference.
6. The employee's credentials were also managed seamlessly by using Azure Service Principal and OAuth Services. 
7. Each graph generation by the model was also accompanied by a summary of the findings from single or multiple graphs.

Impact



The client was able to maintain his data security while being able to create visually appealing graphs and plots and sharing them with external stakeholders and internal employees.

It also increased the productivity of the employees working for the client by letting them create visuals using natural language queries.

The visuals created were shareable throughout the organization using the Microsoft 365 ecosystem. It helped the client to generate better visuals faster and perform more effectively.