Skip to content

CodeSage.md

Latest commit

 

History

History
82 lines (61 loc) · 3 KB

CodeSage.md

File metadata and controls

82 lines (61 loc) · 3 KB

CodeSage — AI-Powered Code Interpreter & Explainer

For full documentation-->[https://Mokshii46.github.io/CODESAGE/] CodeSage is a Python-based framework that combines compiler principles, tree-walk interpretation, and AI-based summarization to not only execute code but also explain it in human-readable form.
It is designed to help developers, students, and educators understand code behavior and structure.

Pipeline Overview

CodeSage processes code through several well-defined stages:

  1. Scanner / Lexer

    • Reads raw source code character by character.
    • Converts it into tokens (keywords, operators, literals, identifiers).
    • Detects lexical errors (unrecognized symbols, missing delimiters).

  2. Parser

    • Applies a recursive descent parser to transform tokens into an Abstract Syntax Tree (AST).
    • The AST represents the logical and hierarchical structure of the code.
    • Detects syntax errors and provides meaningful error messages.

  3. AST Summarizer

    • Traverses the AST to create structured summaries of code behavior.
    • Converts programming constructs (loops, conditionals, assignments) into readable explanations.

  4. Interpreter (Tree-Walk Execution)

    • Executes the AST recursively.
    • Evaluates expressions, executes statements, handles loops, functions, and conditionals.
    • Provides runtime output, including results of calculations and print statements.

  5. NLP Integration (Optional)

    • Takes the AST summaries and refines them using natural language models.
    • Produces human-like explanations of the code’s logic.

  6. GUI / IDE

    • Built with Tkinter, the interface integrates:
      • Code editor
      • Output console
      • AST visualization
      • Summaries and explanations

Results

CodeSage can handle a variety of Python constructs:

  • Variable declarations and assignments
  • Arithmetic and logical operations
  • Loops (for, while) and conditionals (if, elif, else)
  • Functions, return statements, and built-in utilities (len, range)
  • Lists and index-based operations

Example snippet:

x = 0
while x < 5:
    print(x)
    x = x + 1

Problems Faced

  • Lack of suitable dataset for NLP training:
    • Initially, no proper code-summary dataset was available.
    • Had to create a custom template-based dataset and filter it according to the interpreter’s capabilities.
    • Even after training, the NLP output lacked accuracy, so the system shifted toward AST-based summarization.

Future Prospects

  • Advanced NLP Integration: Use pre-trained models like CodeT5 or LLaMA for more natural explanations.
  • Extended Language Features: Classes, modules, and advanced Python constructs.
  • Web-Based IDE: Replace Tkinter with a modern web interface for better usability.
  • Real-Time Code Explanation: Summarize code dynamically as users type.

Mentors

  • Yadnyesh Patil
  • Rupak Gupta

Contributers

  • Mokshi Shah