Every design student has Claude, GPT, Gemini.
They paste their concept, ask "Is this coherent?"
— and get an answer that makes them worse at design.

Koher doesn't tell you what's wrong.
Koher shows you the shape of what you've written.

Training wheels, not autopilot.

The Problem

Why not just use Claude?

Claude Koher
Answers Surfaces what you're not seeing
Hides doubt ● Present · ◐ Unclear · ○ Absent
Creates need Builds your eye

Claude makes you feel smart.
Koher helps you see clearly.

Two Tools

Use one, or both.

Fog → Concept

For students who have fog instead of a concept.

5 questions. 1 exercise.

Static site. No ML.

Concept → Diagnosis

For students who have something written.

5 dimensions. 3 severity levels. Plain diagnosis.

DeBERTa + Haiku API.

Tool 1 ──handoff──→ Tool 2

The Honest Part

Three states, not two.

Your Concept

"My project creates a platform for communities to share local knowledge and preserve cultural heritage through collaborative storytelling."

CLAIM ◐ Unclear
EVIDENCE ○ Absent
SCOPE ○ Unbounded
ASSUMPTIONS ○ Hidden
GAPS ● Connected

When Koher shows ◐ Unclear, it's telling you: "I can detect something here, but your writing is too vague to be sure."

This is more useful than a false positive.

The Framework

Five dimensions of coherence.

Under the Hood

AI handles language. Code handles judgment.

Show architecture
Student text ──→ DeBERTa ──→ Rules ──→ Haiku
                 (98.38%)    (code)    (API)
  • Stage 1: Classification — confidence scores (0.0–1.0)
  • Stage 2: Deterministic rules — severity levels
  • Stage 3: Plain diagnosis — narrates, doesn't judge

Students don't lack information.
They lack attention.

Prayas Abhinav

Prayas Abhinav

Design leader. 15+ years. Product, education, art.

TED India Fellow · Ex-CDO · NID/IIT Faculty · 50M+ viewers (Mumbai Airport T2)

prayas.in ↗ linkedin ↗