Audix Project Story

Inspiration

As software developers, we’ve repeatedly faced the frustration of code not matching documentation, assigned tasks, or company standards. Hidden bugs, missing features, and non-compliant code often caused delays, extra costs, and security risks.

We realized that while there are tools like static analyzers or GitHub Actions, no tool fully automates verification that code actually implements what was planned. This gap inspired us to create Audix, an AI-powered desktop app that audits code against tasks, documentation, and company standards before launch.

What We Learned

Working on Audix has taught us several critical lessons:

1. Data alignment matters more than code quality alone – a perfectly written function is useless if it doesn’t meet the project specification.
   
2. AI can accelerate verification – natural language processing and code analysis together can detect mismatches automatically.
   
3. User workflows must be intuitive – integrating auditing without disrupting developers’ workflow is key to adoption.
   

We also gained hands-on experience with PyQt5 for desktop apps, the Gemini API for AI code analysis, and managing real project data for testing.

How We Built It

Audix is structured around three main components:

1. Desktop Interface

   • Built with Python + PyQt5, enabling cross-platform GUI development.
     
   • Provides a dashboard to view projects, categories of checks, and audit results.

2. AI Verification Engine

   • Uses the Gemini API to analyze source code and documentation.
     
   • Performs task-code alignment: for each task ( t_i ) in project documentation, it checks whether corresponding code ( c_i ) satisfies requirements.
     

Mathematically:

[ \text{Match}(t_i, c_i) = \begin{cases} 1 & \text{if } c_i \text{ implements } t_i \ 0 & \text{otherwise} \end{cases} ]

The system computes an overall compliance score:

[ \text{Compliance} = \frac{\sum_i \text{Match}(t_i, c_i)}{N} \times 100\% ]

1. Project Management Integration
   
   • Links tasks, code, and documentation within one dashboard.
     
   • Provides audit summaries, highlighting mismatches and potential compliance risks.

Challenges We Faced

1. Mapping code to tasks Automatically understanding which code implements which task is non-trivial, especially when variable naming or code structure is inconsistent.
   
2. Balancing AI precision and speed Initial AI checks were slow; optimizing the queries and caching results was essential.
   
3. Cross-project variability Different projects follow different documentation styles. Training Audix to handle diverse structures required iterative testing.
   

Outcome

After nearly a year of development:

• Audix can fully audit a Windows project, checking code against tasks and documentation.
  
• We’ve validated the AI engine with real-world projects, identifying hidden mismatches and potential bugs before launch.
  
• Next steps include beta testing with users, and adding macOS and Linux support.

Built With

• gemini
• pyqt5
• python