Tiny Volt: AI harness for prototyping, scientific publishing and interactive articles

Tiny Volt provides a unified platform to:

  • Publish

    Use Markdown to write interactive articles and textbooks.

    Start writing →
  • Read, annotate and prototype

    Get more out of a PDF by bridging the gap between reading and implementing.

    Upload a PDF →
  • Experiment

    Access to Python's scientific computing toolkit allows readers to modify and run code themselves.

    See examples ↗
Self-host with pie
  • Built-in Python editor
  • Reactive canvas
  • Autodiff
  • Embed in your own site
  • Generative UI
  • 100% open source
npx @tinyvolt/pie init my-site
Visit pie.tinyvolt.com

Examples

All Python code runs in your browser.

k-means clustering

Splits points into kk groups. It repeats:

  1. Color each point by its nearest center.
  2. Move each center to the middle of its points.

Run this first. Rerun it for new points and new starting centers.

pie.clear()
pie.show_canvas()
pie.scalar(0, out='gridOpacity')
import numpy as np
k = 3
gray = '#a8a29e'
colors = ['#e11d48', '#2563eb', '#d97706', '#7c3aed', '#0d9488', '#65a30d']
blobs = [(-2.5, 1), (2.5, 1.5), (0, -2)]
rng = np.random.default_rng()
coords = np.concatenate([rng.normal(b, 0.8, size=(12, 2)) for b in blobs]).clip(-4, 4).round(2)
ps = [pie.point(x, y, fill=gray) for x, y in coords]

# The centers start on k different points, picked at random.
cs = coords[rng.choice(len(coords), k, replace=False)]
centers = [pie.point(x, y, hidden=True) for x, y in cs]
for c, color in zip(centers, colors):
    pie.circle(c, 0.2, fill=color, stroke='#1c1917')
tick = pie.scalar(0)  # not drawn; animating it is a half-second pause

def pause():
    pie.translate_scalar_by(tick, 1, animate=True)

def show(r, changed, spread):
    done = ', done' if changed == 0 else ''
    pie.text(f'round {r}: {changed} changed color, spread {spread:.1f}{done}', 0, 4.8, 14, stroke='#1c1917', out='status')

Then run this.

labels = np.full(len(coords), -1)
for r in range(1, 100):
    # Each point takes the color of its nearest center.
    d = ((coords[:, None, :] - cs[None, :, :]) ** 2).sum(axis=2)
    new = d.argmin(axis=1)
    changed = int((new != labels).sum())
    labels = new
    for c in range(k):
        group = [ps[i] for i in np.flatnonzero(labels == c)]
        if group:
            pie.set_fill(group, colors[c])
    show(r, changed, d[np.arange(len(coords)), labels].sum())
    if changed == 0:
        break
    pause()
    # Each center moves to the average of its points. A center with no points stays put.
    cs = np.array([coords[labels == c].mean(axis=0) if (labels == c).any() else cs[c] for c in range(k)])
    pie.translate_point_to(centers, cs[:, 0].tolist(), cs[:, 1].tolist(), animate=True)
    pause()

A bad start can end in a worse split. Set k up to 6.