MLaaS: Streamlit-Based ML Deployment Platform

MLaaS demonstrates how Streamlit can be used to turn a trained ML model into a deployable, interactive web application — without a full backend stack. The platform makes exploratory data analysis and model inference accessible through a browser, lowering the barrier between a trained model and an end user.

What's Deployed

The live demo at eda.da-tu.ca hosts an interactive EDA (Exploratory Data Analysis) environment built on top of the DaTu data infrastructure. Users can:

• Upload datasets and explore distributions, correlations, and summary statistics interactively
• Run inference with deployed models directly in the browser
• Visualize model outputs and evaluation metrics without writing code

Why Streamlit for ML Deployment

Traditional ML deployment requires: a REST API (FastAPI/Flask), a frontend (React/Vue), hosting infrastructure, and authentication. For internal tools, research demos, and stakeholder-facing prototypes, this is overkill.

Streamlit collapses the stack: one Python file produces a deployable web app. The tradeoffs are real (limited customization, Python-only) but for the primary use case — making ML results interactive and shareable — they're acceptable.

Architecture

The platform runs on a lightweight cloud instance. The Streamlit application connects to the same data pipeline used by the DaTu research platform, pulling processed datasets on demand. Model artifacts are stored as pickled scikit-learn or ONNX files and loaded at inference time.

This project demonstrates a pragmatic approach to ML deployment: match the tool to the use case. A Streamlit app is the right deployment target for a research prototype or internal analytical tool — a full production service is the right target when you have users, SLAs, and reliability requirements.