ML-Projects | CodeForge Portfolio

AI/ML Engineer Portfolio
Python | Scikit-learn | PyTorch | Hugging Face | RAG
Targeting mid-level roles in Japan (Human Resocia / Pasona / BizReach)

---

✅ Completed Projects

1. Japanese RAG Production System

Status: ✅ COMPLETED & PORTFOLIO-READY

Key Achievements

• Built a modular production-oriented RAG pipeline specialized for Japanese documents
• Implemented real RAGAS evaluation (Faithfulness 0.96, Answer Relevancy 0.76) using actual system outputs
• Developed FastAPI backend + Streamlit frontend with clear separation of concerns
• Japanese-aware chunking + bge-m3 embeddings
• Full Docker support + professional Japanese technical summary

Repository: Japanese_RAG_Production

---

2. Credit Card Fraud Detection

Status: ✅ COMPLETED & PRODUCTION-READY

Key Achievements

• High-recall XGBoost (recall 0.92, PR-AUC 0.85)
• SHAP explainability (V14/V17 main drivers)
• Docker container + Streamlit live demo
• Unit tests + pinned dependencies
• Japanese summary + business insights

Live Demo (Streamlit Cloud): [https://ml-projects-credit-card-fraud-detection.streamlit.app/]

---

3. Japanese Sentiment Analysis (NLP)

Status: ✅ COMPLETED & PORTFOLIO-READY

Key Achievements

• Fine-tuned Japanese BERT (cl-tohoku/bert-base-japanese-v2) with 3-class sentiment
• Production deployment on Gradio + Streamlit Cloud (CPU-optimized)
• Model pushed to Hugging Face Hub (Retro099/japanese-sentiment-analysis-v1)
• Professional assets: confusion matrix, Japanese summary

Live Demo: Gradio → https://f50c787d7b105f7bf9.gradio.live/
Streamlit Cloud: [https://cx7v54eehcppwnarlaplxt.streamlit.app/]
Model on HF Hub: https://huggingface.co/Retro099/japanese-sentiment-analysis-v1

---

4. Customer Churn Prediction

Status: ✅ Completed & Live
Tech: Scikit-learn, Streamlit, Pandas

• End-to-end ML pipeline with production-ready artifact
• Interactive Streamlit web application
• Strong business insights and detailed Japanese documentation

Live Demo: Streamlit App

---

All projects follow PEP8 standards, modular structure, and pinned dependencies.
Every project includes a Japanese summary and clear business impact section.

日本就業に向けたポートフォリオ概要
日本でのデータサイエンティスト / MLエンジニア就業を目指してポートフォリオを強化中。在留資格取得手続き中、日本語はN4レベル（N3勉強中）。初回面談は英語メインで対応可能。

特に日本語文書向けRAGシステム、Docker本番運用、SHAP説明性、日本語BERTファインチューニングを強みとするプロジェクト群です。