277

No CodeML

A distributed MLOps platform for client-side data processing, cloud-scale training, and one-click model deployment.

No CodeML is a distributed MLOps platform that takes a dataset from raw CSV to a production inference API without writing a line of code. It was built as my final-year capstone at FAST NUCES and is designed around three ideas: do the heavy work where it makes sense, train at cloud scale, and deploy instantly.

Highlights

  • Distributed Ray pipeline — Orchestrated end-to-end ML workflows on Ray.io, using Ray Clusters for parallelized training and Ray Serve for scalable, auto-scaling model serving across distributed nodes.

  • Client-side processing with WebAssembly — Engineered browser-based data-processing kernels in WebAssembly and JavaScript Workers, enabling parallel parsing and feature engineering of gigabyte-scale CSV files on the client at near-native speed.

  • One-click deploy — A deployment system that transitions a trained model into a production API on Ray Clusters in a single click, with high availability and auto-scaling inference.

Workflow

  1. Data Preprocessing — Upload a dataset, preview it instantly, and apply transformations — all processed client-side via Wasm.
  2. Model Training — Train at cloud scale on distributed Ray Clusters and inspect results.
  3. Model Deployment — Promote the trained model to a live, auto-scaling inference endpoint.

Tech Stack

Python · React · Ray.io · WebAssembly (Wasm) · AWS