PLCSTM: A Deep Hybrid Architecture for Polyglot Source Code Classification

Description

  PLCSTM (Programming Language Classifier using Spatiotemporal Modeling) is a specialized machine learning framework designed to automatically categorize source code into 22 distinct programming languages. Unlike traditional classifiers, PLCSTM is a unified, high-dimensional neural architecture that treats code as a structured sequence of spatial motifs and temporal patterns.

    The model is engineered to extract "code-signatures"—the unique way different languages nest logic and use symbols. This allows the system to differentiate between syntactically similar languages like C# and Java by analyzing the spatiotemporal relationships between tokens. Integrated into a secure Django-based platform, PLCSTM provides an end-to-end "IDE-integrated" experience for automated code labeling and security auditing.

Input

1. Source Snippets: Raw source code text provided via the web interface.
2. Development Scripts: Individual source files or complex logic blocks for real-time identification.

Output

1. Classification: The identified programming language (e.g., Python, C++, Rust, Go, etc.).
2. Confidence Score: A probability percentage (0-100%) reflecting the model’s certainty in its prediction.
3. Syntax Mapping: Metadata used to resolve overlaps in similar web-technology languages.

Key Features

1. Spatiotemporal Pattern Recognition: PLCSTM uses a proprietary layer configuration to identify both local syntax "stamps" (spatial) and the sequential flow of logical execution (temporal).
2. Hybrid Heuristic Refiner: A specialized logic gate that double-checks the neural output against "Smoking Gun" syntax markers (like type annotations or structural braces) to ensure perfect accuracy for CSS and TypeScript.
3. High-Capacity Vocabulary: Utilizes a 15,000-word specialized tokenizer designed specifically for programming lexicons and character sequences.
4. Singleton Execution Engine: Optimized for production; the PLCSTM engine loads into memory once upon deployment, ensuring sub-second inference for every request.
5. VS Code Ecosystem Aesthetics: A fully realized user interface mirroring the Visual Studio Code environment, providing developers with a familiar workspace for code analysis.
6. Secure User Management: A robust Authentication Layer (Login, Register, Logout) that protects the inference engine and manages user-specific history.

Tags

#PLCSTM #SourceCodeClassification #MachineLearning #SpatiotemporalModeling #NLPforCode #Django #AIArchitecture #CodeAnalysis #PolyglotClassifier #VSCodeTheme

References

1.Keras Documentation (2025). Serializing custom subclassed models with register_keras_serializable.

2.Django Project (2025). Implementing secure authentication and singleton asset loading.

3.Spatiotemporal Research. Methods for sequence classification in high-dimensional text data.