Inspiration
My kid was born recently and nurturing a child's imagination is a big challenge that all families face, yet many families and educators face challenges creating personalized, engaging, and educational children's content and I wanted to give him the ability to read and hear stories from his grandparents even when they are not close to us. I want to find ways to convert their memories and the family moral stories that are passed down through generations to my kid in a friendly readable book
What it does
The project takes inputs of a short text of the stories and passes the information through multiple agents to generate a small kid's story book with a cover image. it ensures proper content safety filters are setup to create a cohesive kids story book.
How we built it
I leveraged Github copilot and microsoft copilot in tandem to build this project. I acted as the product owner, solutions architect and QA while the LLM (openai o3mini, claude sonnet 3.7 and gemini flash) all were used to review and build the application. The app uses python flask api, Azure foundry and azure open ai service to deploy the LLM, Azure Agent ai framework, html and css.
- Asynchronous story processing
- Content filtering for child-appropriate material
- DALL-E 3 illustration generation
- HTML story output with downloadable option
- Data persistence in Azure SQL
- Production-ready Azure deployment configuration
Challenges we ran into
Numerous, I am a developer who is just starting out. My Career has been in Databases and Security and there was multiple issues understanding errors and choosing right frameworks was a challenge. Getting the LLM to retain history of changes was also a problem i ran into.
Asynchronous agent calls and combining them was a big challenge LLM Agents like DALL-E-3 has makes spelling errors Deploying to Azure Webapp
Accomplishments that we're proud of
I am proud of being able to develop this app from scratch by iterating over multiple times and juggling a full-time job and a 20-month-old baby. I have huge plans to continue building this project and hope to fully production this in future
What we learned
Azure AI foundry makes it easy to test agents and prompts and coordinate deployments of models required for the project, GitHub copilot makes coding so much fun and easier especially for people who are not traditional developers, it empowers me to bring my ideas to life with little historical coding experience. I am comfortable talking to the AI and how to get it to answer my questions correctly
What's next for EPIK
My plan is to continue developing this further, planning to incorporate multiple features like Getting Audio inputs and supporting other natural languages. I will continue to develop this product further
Built With
- azure
- azureaifoundry
- azuresql
- copilot
- flask
- github
- python
- sqlalchemy
Log in or sign up for Devpost to join the conversation.