InstaML is a no-code machine learning workspace built with React and FastAPI. It lets you upload datasets, run preprocessing pipelines, inspect statistical charts, train classification/regression models, and execute predictions.
InstaML splits Gateway orchestration, Optuna training threads, and Predictor inferences into separate services:
graph TD
Client["React SPA (Client)"] -->|"HTTP / JWT"| Gateway["Gateway API (Port 8000)"]
Gateway -->|"HTTP / RPC"| Trainer["Trainer Service (Port 8001)"]
Gateway -->|"HTTP / Streams"| Predictor["Predictor Service (Port 8002)"]
Gateway -->|"SQL Queries"| DB["SQLite (Local) / Turso (Cloud)"]
Trainer -->|"SQL Queries"| DB
Predictor -->|"SQL Queries"| DB
Gateway -->|"Read / Write"| LocalCache["Local Cache (backend/storage)"]
Trainer -->|"Read / Write"| LocalCache
Predictor -->|"Read / Write"| LocalCache
LocalCache -->|"REST Sync"| Supabase["Supabase Storage Bucket"]
Explore the specialized guides below to configure, run, or deploy the platform:
- Installation & Setup Guide - Run backend microservices and frontend clients locally with or without Docker.
- Production Deployment Guide - Host databases on Turso, objects on Supabase, backend on Render, and frontend on Vercel free-tiers.
- System Architecture Details - Learn about microservice routers, database schemas, and local-cloud cache sync loops.
- Contributing Guide - Find repository directory layouts, design rules, and unit/integration testing logs.
- Intelligent Data Upload: Upload CSV, Excel, Parquet, and ZIP files with automatic column and task detection.
- Interactive Preprocessing: Clean data, handle missing values, drop columns, scale features, and encode categories.
- Exploratory Data Analysis (EDA): Descriptive statistics, distribution plots (Box plots, Histograms, KDE, CDF), correlations, and PCA/K-Means clustering.
- Model Training: Train machine learning models (XGBoost, Random Forests, Multi-Layer Perceptrons) with Optuna hyperparameter tuning and K-Fold cross-validation.
- Specialized Modalities: Pretrained pipelines for text sentiment/summarization/NER, computer vision face detection/OCR, and audio noise reduction/ASR.
- Deployment: One-click model deployment, REST API generation, real-time prediction forms, and dataset version checkpoints.
Start the entire stack using Docker Compose:
- Make sure Docker Desktop is running.
- Launch the services:
docker-compose up --build
- Open http://localhost:5173/ in your browser.
This project is licensed under the MIT License.