Inspiration

Debugging is one of the most time-consuming parts of software development, especially for students and developers who are still learning to identify the root causes of errors. We wanted to build a tool that makes debugging more accessible by combining AI-powered code analysis with actionable explanations and suggested fixes.

What It Does

DevShield AI is an AI-powered Python debugging assistant designed to help developers understand and resolve coding errors.

Its key capabilities include:

Bug Detection: Analyzes Python code to identify potential bugs and coding issues.
Root Cause Analysis: Explains why an error may occur in understandable language.
Code Fix Suggestions: Recommends corrected code to help developers resolve issues.
Unit Test Generation: Generates relevant test cases to help validate code behavior.
Sandboxed Verification: Designed to support safer code execution and verification in an isolated environment.

How We Built It

DevShield AI combines a web-based interface with a Python backend. The frontend uses React, Vite, and Tailwind CSS, while the backend uses FastAPI and Python.

The project is designed to integrate NVIDIA Nemotron through Nebius Token Factory for AI-powered code analysis. The application architecture separates the user interface, backend processing, AI inference, and code execution components.

Challenges We Faced

Building an AI debugging assistant involves more than generating code suggestions. The application must handle different error types, parse model responses reliably, present explanations clearly, and distinguish actual analysis results from failures.

Integrating the frontend with the backend and preparing the application for deployment also introduced challenges around API connectivity, configuration, and reliable error handling.

What We Learned

Through this project, we explored AI-assisted software development, prompt engineering, API integration, full-stack application development, and deployment workflows. We also learned the importance of structured model outputs, transparent error handling, and safe execution practices when building developer tools.

What's Next

We aim to improve debugging accuracy, expand support for different Python error patterns, strengthen sandbox isolation, and evaluate generated fixes through automated tests. Future improvements could include repository-level analysis and a more comprehensive debugging workflow.

Built for the Hackathon

DevShield AI was developed for the Nebius × NVIDIA Global AI Hackathon, under the Coding and Agentic Engineering track, with the goal of making AI-assisted debugging more practical and accessible.