Scribd
Upload
24 views2 pages

AI Code Generation Abstract

This research addresses the limitations of existing AI-driven code generation tools by proposing a hybrid model that combines transformer architectures with reinforcement learning to improve code efficiency, readability, and correctness. A comparative review of ten state-of-the-art models highlights their methodologies and shortcomings, leading to the development of a novel model trained on curated GitHub repositories. Evaluation results demonstrate superior performance in code generation tasks, making it suitable for integration into development environments for real-time intelligent code suggestions.

Uploaded by

srmaharana.1999
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
24 views2 pages

AI Code Generation Abstract

This research addresses the limitations of existing AI-driven code generation tools by proposing a hybrid model that combines transformer architectures with reinforcement learning to improve code efficiency, readability, and correctness. A comparative review of ten state-of-the-art models highlights their methodologies and shortcomings, leading to the development of a novel model trained on curated GitHub repositories. Evaluation results demonstrate superior performance in code generation tasks, making it suitable for integration into development environments for real-time intelligent code suggestions.

Uploaded by

srmaharana.1999
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd

Abstract

Problem Statement:

With the growing demand for rapid software development, AI-driven code generation tools have

gained popularity.

However, existing models like OpenAI's Codex and Google's AlphaCode often struggle with code

efficiency, readability,

and syntactical correctness. Moreover, the lack of contextual understanding and high computational

demands hinder

their real-world applicability. There's a need for a more efficient, context-aware, and accurate model

that bridges

the gap between generation quality and resource optimization.

Objective:

The objective of this research is to propose and evaluate a hybrid AI model for code generation that

integrates

transformer-based architectures with reinforcement learning. The aim is to enhance code

completion, bug detection,

and execution efficiency while maintaining low computational cost and syntactical integrity.

Abstract:

The increasing complexity of programming tasks has accelerated the development of AI-powered

code generation systems.

This study presents a comprehensive comparative review of ten state-of-the-art models, analyzing

their methodologies,

datasets, evaluation metrics, and limitations. To overcome identified gaps, we propose a novel

hybrid model that


combines transformer-based architectures with reinforcement learning and AST-aware encoding.

The model is trained using

curated GitHub repositories and fine-tuned for optimal performance in code generation and

correction tasks. Evaluation

results show superior performance compared to GPT-3 and AlphaCode in BLEU score, accuracy,

execution time, and perplexity.

The system is suitable for integration into development environments through APIs or IDE

extensions, enabling real-time,

intelligent code suggestions. This research provides a robust foundation for next-generation code

assistants by enhancing

contextual understanding, efficiency, and correctness.

References:

1. Almorsi, A., Ahmed, M., & Gomaa, W. (2025). Guided Code Generation with LLMs: A Multi-Agent

Framework for Complex Code Tasks.

arXiv preprint arXiv:2501.06625.

2. Lu, S., Wang, D., et al. (2023). CodeT5+: Open Code Large Language Models for Code

Understanding and Generation. ACL.

3. Li, Y., Qiu, X., et al. (2023). AlphaCode: Using LLMs for Competitive Programming. Nature.

4. Feng, Z., Guo, D., et al. (2020). CodeBERT: A Pre-Trained Model for Programming and Natural

Languages. EMNLP.

You might also like