AI-Hallucination-Detector-for-Code

This is a submission for the GitHub Copilot CLI Challenge

What I Built

I built AI Hallucination Detector for Code, a command-line tool that detects hallucinations in AI-generated Python code using static analysis.

The tool identifies:

• Hallucinated or fake APIs (non-existent imports)
• Logical inconsistencies such as misleading time-complexity claims
• Produces a clear hallucination risk score to help developers judge reliability

This project matters to me because AI-assisted coding is becoming mainstream, but correctness and trust remain critical. Instead of generating more code, this project focuses on verifying AI-generated code responsibly.

---

Demo

🔗 GitHub Repository:

https://github.com/anupam-hegde/AI-Hallucination-Detector-for-Code

🎥 Demo GIF / Video:

(Attached in the repository README)

Demo Flow:

1. A Python file with a fake import and nested loops is analyzed
2. The CLI detects:
   • API hallucinations
   • Logic hallucinations
3. A risk score is generated with a clean, readable CLI output

The demo shows the tool working end-to-end using a single CLI command.

---

My Experience with GitHub Copilot CLI

GitHub Copilot CLI played a key role as a development and reasoning assistant throughout this project.

I used Copilot CLI to:

• Understand and implement AST-based static analysis
• Design safe logic for validating Python imports without executing code
• Debug CLI architecture issues related to Typer
• Reason about algorithmic complexity and scoring strategies

Copilot helped me move faster and think more clearly, but it was never treated as a source of truth.

All suggestions were manually reviewed and validated using deterministic static analysis — which directly aligns with the goal of this project.

This experience reinforced how Copilot CLI can be a powerful productivity tool when used responsibly and thoughtfully.

---