A serverless, FinOps-optimized data lakehouse built on AWS - ingesting real-time and batch data, transforming it through a Medallion architecture, and delivering business insights through an intera…

This CRAFT + Llama 3.1 pipeline automates invoice semantic extraction, elevating accuracy from 65% to 94%. It achieves a 92% success rate in line-item reconstruction and is deployed via Flask API t…

This project uses NLP to classify fake news headlines (GossipCop/PolitiFact). It compares the performance of BERT embeddings when fed into tuned Machine Learning (like XGBoost) and Deep Learning (L…

This project models the GitHub collaboration network to predict potential partnerships. It applies Social Network Analysis (SNA) and a Graph Autoencoder (GAE) to tackle the link prediction task, id…

This research focuses on group detection and collaboration analysis among GitHub developers. It builds a commit-based collaboration network, where nodes represent developers and edges indicate join…

This project presents a comprehensive comparison of classical, embedding-based, and hierarchical (hyperbolic) topic modeling approaches on over 57K GitHub README files, highlighting the proposed G…

This project demonstrates a classic OCR pipeline. This Flask app takes an image, applies an OpenCV preprocessing pipeline, and uses Tesseract OCR to digitize Vietnamese invoices (Bách Hóa Xanh)..

This project transforms messy invoice images into a structured, searchable knowledge base. The pipeline automatically extracts text with Tesseract, uses Google Gemini to parse fields (vendor, total…

This project automatically classifies GitHub repositories using README.md text. It combines classical ML models with fine-tuned Transformers (Mistral-7B + PEFT/LoRA) on a large dataset of 50+ IT to…

An unsupervised clustering analysis of over 57,000 GitHub projects using their README.md text. This study compares traditional keyword methods (TF-IDF) against modern semantic embeddings (Sentence-…

A collection of projects for the "Statistical Machine Learning" course. Demonstrates a wide range of techniques from classic Scikit-learn models (KNN, SVM, KMeans) to Statsmodels (ARIMA) and Deep L…

This is my programming 2 project, building a human resource management software using