2025ML Engineer (Solo)Archived

Symptom Checker AI

A machine learning healthcare assistant that predicts likely conditions from reported symptoms.

Source code

92%+

Accuracy

scikit-learn

Model

ML / AI

Type

Problem

Turning a list of self-reported symptoms into a useful, non-alarming set of possible conditions is a classification problem with real stakes: the model needs to be accurate enough to be useful and clear enough that users understand it's a starting point, not a diagnosis.

Approach

Trained and evaluated multiple scikit-learn classification models against a labeled medical symptom dataset.
Applied feature engineering to map free-form symptom input onto the model's expected feature set.
Exposed the trained model through a Flask REST API, with a simple frontend for entering symptoms and reviewing predictions.

Key decisions

scikit-learn over a deep learning model

The dataset size and feature structure suited classical ML better than a neural network, and kept the model interpretable and fast to iterate on.

Flask API boundary

Separating the model behind a REST API keeps the ML pipeline swappable without touching the frontend.

Tech stack

PythonMachine Learningscikit-learnFlask

Outcomes

92%+ prediction accuracy on the evaluation set

Back to home