retro099/ml-projects

How retro099/ml-projects works

This portfolio contains four independent ML demos. Users submit transaction features, customer details, Japanese text, or a document question and receive a prediction or grounded answer. The RAG project has the richest workflow: document preparation and Japanese-aware chunking feed vector retrieval, then an API coordinates retrieval and answer generation; a Streamlit interface is also documented. Fraud and sentiment load pretrained models, while churn trains its baseline from sample data at startup. Some RAG ingestion and UI wiring is documented but not shown in source excerpts, so those relationships are cited to the README or left unverified.

Main entry points: api.py Ask API · embedding_store.py Embedding Store · streamlit_app.py Streamlit UI · ingestion.py Document Ingestion

Components 20

Japanese Document Q&A

Streamlit UI Japanese_RAG_Production/app/streamlit_app.py
Ask API Japanese_RAG_Production/api.py
Company PDFs
Document Ingestion Japanese_RAG_Production/src/ingestion.py
Japanese Chunking Japanese_RAG_Production/src/chunking.py
Chunk Import Japanese_RAG_Production/ingest_chunks_jsonl.py
Embedding Store Japanese_RAG_Production/src/embedding_store.py
Retriever Japanese_RAG_Production/src/retrieval.py
Answer Generator Japanese_RAG_Production/src/generation.py
Groq LLM

Credit Card Fraud

Fraud Demo Credit_Card_Fraud_Detection/app.py
Fraud Model Credit_Card_Fraud_Detection/artifacts/fraud_model_v1.joblib

Customer Churn

Churn Demo Customer_Churn_Prediction/app.py
Sample Customers Customer_Churn_Prediction/data_sample/Customer_Churn.csv

Japanese Sentiment

Sentiment Demo Japanese_Sentiment_Analysis/app.py
Hugging Face Model

Other components

Questioner
Transaction User
Churn User
Text User

Connections 16

Questioner → Streamlit UI: asks question README.md
Streamlit UI → Ask API: submits question README.md
Ask API → Retriever: retrieves context Japanese_RAG_Production/api.py
Ask API → Answer Generator: requests answer Japanese_RAG_Production/api.py
Retriever → Embedding Store: queries vectors Japanese_RAG_Production/src/retrieval.py
Answer Generator → Groq LLM: generates answer Japanese_RAG_Production/src/generation.py
Company PDFs → Document Ingestion: provides documents README.md
Document Ingestion → Japanese Chunking: chunks text README.md
Japanese Chunking → Embedding Store: embeds chunks README.md
Chunk Import → Embedding Store: adds chunks Japanese_RAG_Production/ingest_chunks_jsonl.py
Transaction User → Fraud Demo: enters features Credit_Card_Fraud_Detection/app.py
Fraud Demo → Fraud Model: loads model Credit_Card_Fraud_Detection/app.py
Churn User → Churn Demo: enters details Customer_Churn_Prediction/app.py
Churn Demo → Sample Customers: reads training data Customer_Churn_Prediction/app.py
Text User → Sentiment Demo: submits text Japanese_Sentiment_Analysis/app.py
Sentiment Demo → Hugging Face Model: classifies text Japanese_Sentiment_Analysis/app.py

More Jupyter Notebook diagrams: terry-cyx/metaopenfoam · wondonghyeon/protest-detection-violence-estimation · dali92002/docentr · jared-chan/f1ml · felisat/clustered-federated-learning

Drawn by GitDiagram on September 30, 2026 from the repository's file tree, README and source files. Read as Markdown ·