Dataset viewer
Browse a dataset and the concern each response should be classified as. Read-only — nothing is sent to a model.
Add a dataset
Upload a CSV and it's saved into the datasets/ folder — ready to view here and run on the Run Tests tab.
- The first row must be the header id,text,labels.
- id — a short identifier (e.g. T01).
- text — the survey response. Wrap it in "double quotes" if it contains a comma.
- labels — the concern(s) it should be classified as, comma-separated. Leave empty for a benign response.
Example:
id,text,labels T01,"My supervisor belittles me in every meeting.",bullying T02,"Great year overall, felt really supported.", T03,"I can't go on and work stress makes it worse.","suicide_risk,mental_health"
Recognised labels (each gets its own colour; any others still work, shown in grey): suicide_risk, mental_health, bullying, sexual_harassment, racism, homophobia, transphobia, stress, excessive_workload
Critical Concern
Run survey responses through the classifier — local model, GPT-4o, or both — and score them against the dataset labels.
Prompt
sent to both engines on every classifyHistory
saved runs (also written as CSV in runs/)Analyse a spreadsheet
Upload an Excel (.xlsx) or CSV export — any columns, any size — and the local model summarises the key trends. It works with whatever it's given.
Settings
Connection, the models installed on the server, and the raw API logs.
Connection
GPT-4o (OpenAI) — second classifier
A cloud model you can compare against the local model on the Critical Concern tab. Both run the same prompt.