Python Cluster Analysis Web App

The problem

Clustering is useful well beyond the people who can code it. The client wanted a tool that would let an analyst who understands clustering techniques, but not Python, load a dataset, try several algorithms, see the structure and take the result away in a spreadsheet, while still being flexible enough for ML engineers. The product was built for Reuben Vandeventer and is owned by SecondSight.

What I built

• A Python web application with a fixed workflow: upload a CSV or XLSX file, choose the record-name column that uniquely identifies each row, optionally normalise the data, and select X, Y and Z lenses as the three dimensions for analysis.
• Three clustering algorithms, DBSCAN, MeanShift and MST, each with its hyperparameters exposed in the UI so users can tune them without touching code.
• An interactive D3 force-directed network graph as the main output, with nodes selectable and coloured by dimension so clusters can be inspected visually.
• Export of the enriched dataset to CSV or Excel with new columns for cluster IDs and other useful information for downstream analysis.
• Deployment of the app on Heroku so the client could use it from a browser without any local setup.

How it was delivered

The whole product runs as a single web app, so the upload, normalisation, clustering and graph rendering live behind one URL. Hosting it on Heroku kept operations simple for a small team: no servers to manage, and the client could share a link rather than install anything. The original Heroku demo has since been retired.

Outcome

The application made clustering accessible to non-coders and ML engineers alike, with the full loop from raw spreadsheet to interactive graph to exported cluster IDs handled inside the browser. SecondSight holds all rights to the product.