Not information.
Understanding.
Every concept in science, math, and medicine — rendered as an animated diagram by a network that rewards the best explanation. Validators score structure, animation, coverage, and accuracy the best diagram wins.
Three roles. One pipeline.
Render
Runs a 4-step pipeline: retrieves Wikipedia facts, builds a visual_plan JSON (stages and flows), renders an animated SVG with CSS @keyframes, then self-validates before submission.
→ emits miner.svg_eventScore
Scores each SVG across 5 dimensions — validity (15%), animation (20%), concept coverage (25%), plan alignment (20%), LLM accuracy judge (20%). Anti-gaming: required concepts must appear as visible labeled elements, not hidden text.
→ emits validator.score_eventServe
The top-ranked output is returned to your application. The structured visual_plan JSON is included alongside for downstream use.
→ emits output.svg_event// No text walls. No summaries. No paragraphs explaining what a diagram should look like. .
AI gives you information. We give you understanding.
Visual reasoning
spatial > textComplex systems — metabolic cycles, molecular cascades, evolutionary trees — have spatial structure. No paragraph can carry that structure. A diagram can.
Competitive quality
n → bestMiners compete on every query. Validators converge. The network doesn't average quality — it surfaces the best representation the network can produce.
No standard exists
0 prior artThere is no canonical visual representation for gradient descent or quantum superposition. Reprisma builds one, concept by concept, scored by consensus.
Self-improving loop
score ≥ 0.75High-scoring SVGs are collected as structured training examples — visual_plan JSON paired with the diagram. The network's floor rises every epoch.
The difference between reading about a process and seeing it move is the difference between information and understanding.
Built for products, not demos.
Concept visualization
Generate a structured animated diagram for any educational concept. Supports explain, trace (step-by-step), compare (side-by-side), and misconception-correction modes.
> reprisma.visualize({ concept: "neural networks", type: "trace" })AI tutoring interfaces
Drop visual explanations into tutoring apps. Instead of returning text, return an SVG diagram the student can see and follow.
> reprisma.explain({ topic: "transformer architecture", level: "undergraduate" })Training data generation
High-scoring SVGs (score ≥ 0.75) are collected as structured training examples — visual_plan JSON + SVG — ready for fine-tuning visual reasoning models.
> reprisma.dataset({ subject: "physics", min_score: 0.75 })Knowledge systems
Build visual encyclopedias or structured documentation tools. Every concept gets a canonical best-representation rather than a wall of text.
> reprisma.diagram({ concept: "quantum entanglement", format: "svg" })One concept in. Pure clarity out.
SVG + JSON output
Every response includes the animated SVG string and the structured visual_plan JSON. Use both downstream or just one.
Anti-gaming validation
Validators check that required concepts appear as visible labeled elements not hidden text, not metadata buried in attributes.
Research-grounded
Miners retrieve facts from peer-reviewed sources and curated academic corpora before rendering. Every output is grounded in verified knowledge not hallucinated structure.
# pip install bittensorimport bittensor as btreprisma = bt.connect(network = "finney",subnet = 47,min_quality = 0.85,)diagram = reprisma.visualize(concept = "neural networks",challenge_type = "trace",subject = "computer science",difficulty = "intermediate",required_concepts = ["backpropagation", "gradient descent"],)# diagram.svg_output → animated SVG string# diagram.visual_plan → structured JSON
The representation gap is the understanding gap.
“The bottleneck isn’t access to answers. It’s access to the right representation of knowledge at the moment you need it. Text rarely provides that. A well-structured diagram almost always does.”
— FOUNDING CONTRIBUTORS