Bug-Zero AI-Powered Code Bug Finder and Fix Suggestion Platform

Author

R. Nithin, R. Navin Chandra, S. Krishna Prasath, Dr. S. Prakash


Abstract

 Finding bugs in source code is a slow task, especially for students and beginners, because compiler messages are often cryptic and traditional linters work for only one language or follow fixed rules. This project presents BUG-ZERO, a web-based application that uses a Large Language Model (LLM) to find bugs in source code written in any programming language. The application combines an HTML, CSS and JavaScript frontend with a Python Flask backend and an SQL database. Users register and log in, paste or upload code into an editor with syntax highlighting, and receive a structured report that lists each bug with its line number, type, severity, a plain-language explanation, and a suggested fix, together with a complete corrected version of the code. An optional problem statement can be supplied; in that case BUG-ZERO lists each requirement, marks it as met or not met, and reports every mismatch as a bug. The backend uses password hashing and JWT-based sessions, sends the code to the Google Gemini API through a structured prompt, retries when the model is busy, and saves every report so that users can search, reopen, and delete their history. The supplied test script runs 47 automated checks with a simulated model response, and all 47 pass. No live-model accuracy measurement is included in the project files, so the quality of the detected bugs depends on the LLM and should be reviewed by the user. BUG-ZERO is intended as a learning and debugging aid, not as a replacement for code review, testing, or compilation.


Keywords

Bug Detection, BUG-ZERO, Large Language Model, Google Gemini, Flask, JWT Authentication, SQL Database, Code Analysis, Web Application.



Full Text:

Download Paper PDF


References


  1. Google, "Gemini API documentation." Available: https://ai.google.dev/gemini-api/docs
  2. Vaswani et al., "Attention Is All You Need," in Advances in Neural Information Processing Systems (NeurIPS), 2017.
  3. B. Brown et al., "Language Models are Few-Shot Learners," in Advances in Neural Information Processing Systems (NeurIPS), vol. 33, 2020.
  4. Chen et al., "Evaluating Large Language Models Trained on Code," arXiv preprint arXiv:2107.03374, 2021.
  5. Jones, J. Bradley, and N. Sakimura, "JSON Web Token (JWT)," RFC 7519, Internet Engineering Task Force, 2015.
  6. Flask, "Flask Documentation." Available: https://flask.palletsprojects.com/
  7. PyJWT, "PyJWT Documentation." Available: https://pyjwt.readthedocs.io/
  8. SQLite, "SQLite Documentation." Available: https://www.sqlite.org/docs.html
  9. CodeMirror, "CodeMirror 5 User Manual." Available: https://codemirror.net/5/doc/manual.html
  10. Project source code and test artifacts, "BUG-ZERO: AI-Powered Code Bug Finder," supplied project archive, BUG-ZERO.zip. Includes backend, frontend, database schema, README.md, docs/problem_statement.md, and tests/check_requirements.py.

 

 

Share your valuable work from Social Media Buttons