Our summary
This project is a search tool that finds specific concepts inside computer code using everyday language. A person would use this to locate where a program handles tasks like logging in, without needing to guess the exact words the original programmer wrote.
The software uses Jev, an AI tool that chooses from options rather than writing an answer, to score how well files match the search. It lowers its passing score in rounds until it finds enough matches, skipping known private files like access codes along the way.
The search tool relies on a private access code to reach the AI provider and counts connection errors instead of hiding them. It is useful for programmers or AI assistants exploring unfamiliar files, though users must test different settings to get the best results.
Key takeaways
- Lower the required matching score in rounds until the tool finds enough complete files.
- Switch to a backup provider only during connection errors, and always count the failed attempts.
- Test different search strategies and file sizes with the included testing tool before changing default settings.