Inspiration
Learning programming can be difficult for beginners, especially when they encounter complex errors that are hard to understand.
Many students and junior developers struggle not because they lack motivation, but because debugging feedback is often unclear and unstructured. Error messages can be confusing and repositories can be difficult to navigate.
VisionTeacher AI was inspired by the idea that debugging should be a learning experience. Instead of simply showing errors, an AI assistant should help developers understand the problem and guide them step-by-step toward fixing it.
The goal of this project is to transform programming errors into structured development tasks that are easier to understand and solve.
What it does
VisionTeacher AI is an AI-powered developer assistant that analyzes programming problems using multimodal inputs.
Users can submit:
• code snippets
• screenshots of errors
• files or documents
• GitLab repository URLs
The system analyzes the input using Google Gemini and generates clear explanations of the problem.
It then converts the analysis into structured GitLab-style issue checklists that guide developers through the debugging process step by step.
How we built it
VisionTeacher AI is built using a lightweight but powerful architecture.
The backend is developed with FastAPI, which handles API requests, file uploads, and communication with external services.
Google Gemini API is used as the AI reasoning engine to analyze code, screenshots, and user inputs.
The system can also connect to GitLab repositories through the GitLab REST API to inspect repository structures and provide context-aware analysis.
The backend combines the repository context with the user input and sends structured prompts to Gemini, which generates debugging explanations and improvement suggestions.
Challenges we ran into
One of the main challenges was designing prompts that produce structured outputs instead of generic explanations.
Another challenge was processing different types of input such as text, files, and images before sending them to the AI model.
Integrating repository information with AI analysis also required careful handling of context so the model could understand the project structure.
Accomplishments that we're proud of
We successfully built a working AI system that can analyze programming problems and convert them into structured development tasks.
The project demonstrates how AI can be integrated into real developer workflows rather than just providing simple text responses.
We are also proud that the system supports multiple input types including screenshots, files, text, and repository URLs.
What we learned
During this project we learned how to design prompts that guide AI models toward structured reasoning.
We also explored how multimodal AI models can help developers understand code problems more effectively.
This project showed how combining AI with developer platforms like GitLab can create powerful tools for debugging and learning.
What's next for VisionTeacher AI
Future improvements include:
• real-time AI coding assistant
• automatic GitLab issue creation
• deeper repository analysis
• voice explanations for debugging
• integration with classroom learning environments
The long-term vision is to create an AI-powered development assistant that helps developers learn programming through guided debugging.
Log in or sign up for Devpost to join the conversation.