See the science differently.
Graphs, diagrams, circuits, tables and equations carry most of what STEM teaches — and almost none of it reaches a screen reader. PRISM takes a visual and gives it back five ways: as prose, as structure, as a navigable picture, as sound, and as data.
Runs entirely in your browser. No account, no upload, no tracking. Honest about what it cannot read.
Most STEM information is drawn, not written.
Visual by default
A distance–time graph, a labelled cell, a circuit loop, a right triangle: the meaning is in the shape, the position and the colour. Alt text, when it exists, usually says “graph” and stops.
Blind and low-vision students
A screen reader can read every word of a physics textbook and still miss the point of the page, because the point is in Figure 3.
Different processing styles
Many people who are not blind still take in a diagram better as a list of parts, a spoken walk-through, or a table of numbers. One representation never fits everyone.
One triangle, five ways.
Drag the sliders. The picture, the description and the numbers all come from the same structured source — this is the same pipeline the converter runs.
Every visual comes back as…
Natural-language description
Structured prose at three levels of detail — “Triangle. Base: 3 cm. Height: 4 cm. Hypotenuse: 5 cm. Right angle: between base and height.” For graphs: axes, units, trend, minimum, maximum, number of series, slope.
Structured representation
Objects → properties → relationships as an ARIA tree you can walk with the arrow keys, plus a plain-text outline you can copy anywhere.
Accessible SVG
The visual redrawn so that every component is focusable and labelled. Tab through the parts, press Enter to ask “What is this?”, switch on high contrast or large labels, and download it.
Audio description and sonification
Read aloud with your browser's speech engine — any component on demand — and hear graphs as sound: pitch follows the data as time sweeps along the x-axis.
Data extraction
Graph and table values as an editable table with slope, intercepts, minimum and maximum, exportable as CSV or JSON.
And one question: “What happens if…?”
Parametric samples — a triangle, a circle, a lever, a circuit, y = a·x² + c, vector addition — let you change a value and hear, see and read the consequence in every representation at once.
What PRISM can and cannot do
It reads structure it can actually see
Semantic SVGs are parsed from their markup — real values, real labels, real relationships. Raster images (PNG, JPG) go through pixel heuristics: axes, grid lines, coloured curves, shapes, and clusters that look like text.
It never pretends to read text
There is no OCR and no machine-learning model in PRISM. Text in a photo or screenshot is reported as “text (not read)”, with a confidence score for the overall classification, and a Refine panel lets you supply the labels yourself.
Read the full account of the pipeline, its limits and the roadmap →
Try it with your own figure.
Drop in a PNG from a worksheet, paste a screenshot of a graph, or start from fourteen samples that cover graphs, geometry, circuits, cells, levers, tables, flowcharts, functions, vectors and molecules.
Convert a visual