codet5

Solving the task of finding source code given its description, which could be useful for both experienced and novice programmers

A research project to fine-tune a Large Language Model (LLM) to generate code from commands in Bahasa Melayu.

Repository for the master's thesis: "Dr. Boot: Bootstrapping Program Synthesis Language Models to Perform Repairing" (van der Vleuten, 2023)

A Machine Learning research project focused on automated Software Vulnerability Classification (CWE) using fine-tuned linguistic models (CodeBERT, RoBERTa, CodeT5) on the Juliet C/C++ Test Suite.

Interactive DSA Practice platform , A full-stack web app for coding practice through personal contest style tests, with real-time code execution and automated code analysis using a fine-tuned CodeT5 model + Gemini API.

Generating Vulnerability Security Fixes with Code Language Models

Knowledge distillation pipeline for training a smaller CodeT5-based model to generate test assertions. Part of a Bachelor Thesis at TU Delft focusing on efficient automated test assertion generation.

This study compares three transformer-based mod- els—CodeT5, CodeBERT, and CodeGen.

Fine-tuning CodeT5 for Python code generation on the MBPP dataset. Features custom TensorFlow training loops, mixed precision, XLA optimization, and distributed multi-GPU strategies.

AI-powered diagnostic assistant that rewrites C/C++ compiler errors into security-focused explanations using a fine-tuned CodeT5 model.

A machine learning-based system for automated assessment of Java code using CodeT5+ embeddings, feature fusion, and CatBoost regression to predict quality scores from 0 to 10.

human understandable explanation for software vulnerability prediction

AI-powered code review comment generator — a CodeT5 model fine-tuned on real GitHub PR reviews, with a fine-tuned vs. baseline ablation comparison. DLNLP mini-project.

Privacy-preserving code completion with AST-based obfuscation and CodeT5

Code Archaeologist is an intelligent developer tool that analyzes legacy and messy codebases to uncover hidden insights — detecting dead code, measuring complexity, and generating human-readable narratives of code evolution using static analysis and NLP.

Fine-tuning and evaluation of CodeT5 (small vs base) for automatic C# to Java code translation, using the CodeXGLUE dataset.