Data tools / Browser demo

Data Insight Advisor

Explore a dataset and see why particular machine-learning models might suit it.

My contribution: I built the browser application, data-profiling workflow and TypeScript recommendation engine.
Tools & methods: TypeScript · DuckDB-WASM · Web Workers

Existing walkthrough of the upload, exploration and recommendation interface. The score explains a recommendation; it does not report predictive accuracy.

The problem

What this project investigates

Choosing a model starts with understanding the data: missing values, distributions, relationships and the task itself. A recommendation is more useful when the reasoning is visible.

The build

What I built

The app accepts CSV, Excel and JSON files, profiles tabular data with DuckDB-WASM, and presents charts and model-family recommendations. Users can inspect the rules behind a suitability score and browse a model encyclopedia.

Technical choices

Decisions behind the build

Keep profiling off the main thread

DuckDB-WASM runs in a Web Worker so profiling does not share the interface thread.

Make recommendations inspectable

A TypeScript rule engine connects recommendations to dataset statistics. The scores are heuristics, not measured model accuracy.

Process data in the browser

The documented design keeps dataset processing local. This review did not independently audit every network request or privacy guarantee.

Current status

What the evidence shows

A compiled browser application and a walkthrough recording are included in the portfolio repository. The underlying source and its claimed test suite are not included here, so those tests were not rerun.

Further work & evidence gaps

A useful next addition is a worked example that compares a recommendation with actual model evaluation.