Projects Portfolio

Practical projects at the intersection of Data, Machine Learning, and AI Systems

Explore Projects

Filter by Technology

🏆 Featured NVIDIA Nemotron Reasoning Challenge — solver-distilled curriculum feeding a LoRA adapter

NVIDIA Nemotron Reasoning Challenge

Kaggle competition on few-shot rule induction across 6 reasoning problem types (bit manipulation, physics gravity, unit conversion, numeral systems, symbol transform, encryption ciphers). Approach: a solver-distilled curriculum of verified teacher examples used to fine-tune NVIDIA's Nemotron-3-Nano with LoRA. The physics-gravity generator was not observed in the public competitor notebooks reviewed.

Built deterministic per-type solvers and generated 21,311 curated training records across 5 of the 6 problem types. Round-trip re-verification caught errors in the synthetic generators. A rank-32 LoRA training pipeline was implemented for the 30B model, targeting Kaggle's vLLM inference backend.

Final score: 0.588 Rank: 3571 / 4182
PythonPyTorchTRLvLLMHuggingFaceLoRA
RAG HR Assistant

CV RAG System – HR Assistant

Learning-focused project exploring Retrieval-Augmented Generation (RAG) for document understanding and question answering. Built an end-to-end pipeline for PDF parsing, embeddings, vector storage, and natural language querying.

PythonLangChainChromaDBStreamlitLLM
Cloud Market Intelligence Agent — scheduled pipeline from cron to Telegram delivery

Cloud Market Intelligence Agent

Freelance project. B2B AI agent generating scheduled weekly and monthly competitive-intelligence digests for the cloud-services market. Pipeline: cron scheduler → Claude API analysis → quality gates (TL;DR, length, citations) → Telegram delivery + GitHub audit trail. 3-layer architecture: analysis logic / Python runtime (scheduler + bot + HTTP daemon) / Docker deployment.

PythonClaude APIDockerFastAPITelegram BotGitHub Actions
Freelance — repository private
Fruit ripeness classification stages: Unripe, Breaker, Turning, Ripe

Fruit Classification (Thesis)

Final year thesis (2020): built a YOLOv4-based computer vision pipeline to classify fruits by ripeness (green vs. ripe) using color analysis. Designed for agricultural sorting simulation. Implemented with Python, OpenCV, and deep learning training from scratch on a custom dataset.

PythonYOLOv4OpenCVDeep LearningComputer Vision
Thesis Project — repository not yet public
Football Prediction

Football Data Analysis & Prediction

Sports data analysis project using historical football match data (2018–2022). Focused on feature engineering and exploratory statistical analysis to understand predictive patterns.

PythonPandasNumPy
AI Teaching Assistant

AI Teaching Assistant

Learning project combining RAG with Large Language Models for educational question answering using Google Gemini API and grounding techniques.

PythonRAGLLMs
COVID Analysis

COVID-19 Global Data Analysis

SQL-based exploration of global COVID-19 datasets, focusing on trends in cases, deaths, and vaccinations with Tableau visualizations.

SQLTableauData Visualization
Tic Tac Toe AI

Tic Tac Toe AI (Minimax)

Classic game AI implementing the Minimax algorithm with Alpha-Beta pruning to explore decision trees and adversarial search.

PythonMinimaxGame Theory
Python OOP Card Game

Python OOP – Card Game

Object-Oriented Programming practice project implementing a classic card game ("War"). Focused on clean class design, inheritance, encapsulation, and game state management.

PythonOOPDesign Patterns

Let's Connect

Open to thoughtful conversations, collaborations, and meaningful work.

Email
nvnp203@gmail.com
LinkedIn
nphuong302
Location
Ho Chi Minh City, Vietnam

Blending Technology, Art, and Ethics — one conversation at a time.