Mimo — Every Diagram Has a Story
Inspiration 💡
Imagine being told to "look at the diagram" when you cannot see it.
For millions of students, diagrams are an essential part of learning science. But for blind and low-vision students, a biology diagram can become a barrier instead of a learning tool.
A screen reader may read the labels. An AI image description may identify the objects. But neither necessarily explains how those objects connect, why they matter, or what the student needs to understand.
Being able to hear a diagram is not the same as being able to learn from it.
That gap inspired Mimo.
We wanted to build something that does more than describe images. We wanted to create a friendly AI learning companion that helps students explore, understand, and connect ideas independently.
What It Does 🔍
Mimo transforms educational diagrams into accessible, interactive learning experiences.
Instead of giving students one long description, Mimo is designed to help them explore a diagram step by step, using clear language, audio, and interactive navigation.
For example, when exploring an animal cell, a student can learn about the nucleus, mitochondria, and other organelles, one concept at a time.
But Mimo's biggest innovation is something we call Gap Check — our Missing Knowledge Detector.
What if the AI description forgets something important?
Imagine an AI describes mitochondria as structures inside a cell but never explains their role in producing usable energy.
A student might recognize the name without understanding the concept.
Gap Check is designed to identify these missing explanations and suggest information that could make the lesson more complete.
Most AI tools ask: "What is in this image?"
Mimo asks: "What is missing from this explanation?"
Our goal is to make accessibility about understanding, not just description.
How We Built It 🛠️
We designed Mimo using Google's Gemini AI models and an accessibility-first approach.
Our system follows a simple pipeline:
Educational Diagram → AI Description → Gap Check → Verified Explanation → Accessible Lesson
- Understand the diagram: Analyze educational visuals and their existing descriptions.
- Build the lesson: Organize information into small, understandable explanations.
- Find missing knowledge: Gap Check identifies potentially omitted concepts or relationships.
- Check suggestions: Compare proposed additions with available diagram evidence and learning context.
- Support exploration: Present the information through a student-friendly learning experience.
We focused on keeping explanations understandable rather than overwhelming students with technical vocabulary.
Testing Our Innovation
We conducted an initial experiment using Gemini 3.5 Flash-Lite on four biology diagrams.
For each diagram, we compared its original description with a deliberately shortened version.
Our question was simple:
If important information is removed, will Gap Check notice?
Our results:
| Description | Accepted Gap Check Suggestions |
|---|---|
| Original descriptions | 6 |
| Shortened descriptions | 21 |
When we removed information, Gap Check produced more accepted suggestions.
This is encouraging early evidence that the system responds to missing explanations.
However, our experiment is small. Human evaluation is still needed to determine whether the suggestions are factually correct, educationally useful, and genuinely helpful to students.
Challenges We Faced 🚧
1. Describing something is not the same as explaining it.
A diagram contains labels, relationships, processes, and meaning. Teaching those relationships requires more than identifying objects.
2. How do we know when information is missing?
This became our central technical challenge. We explored how AI could compare descriptions with diagram content and identify potentially important omissions.
3. AI suggestions are not automatically correct.
A model may suggest additional information that sounds useful but is inaccurate, unnecessary, or unsupported. We therefore treated verification as an essential part of the design.
4. Making complex science feel simple.
We wanted Mimo to feel like a patient learning companion, not a complicated scientific tool.
What We Learned 🌱
Our biggest realization was:
A longer description does not necessarily create a better learning experience.
Students need explanations that connect ideas, build understanding, and provide the right information at the right moment.
We also learned that measuring AI quality requires more than asking whether its answers sound good.
We need to investigate what the model misses, what it adds, and whether those additions improve understanding.
Our early Gap Check experiment gave us a starting point for answering those questions.
What's Next? 🚀
We want to:
- Test Mimo with blind and low-vision students and accessibility experts.
- Expand from biology to chemistry, physics, and other STEM subjects.
- Improve Gap Check's accuracy and evidence verification.
- Measure whether students understand concepts better after using Mimo.
- Develop more personalized, interactive learning experiences.
Our vision is a future where students are not excluded from understanding a concept simply because it was presented visually.
Why Mimo Matters ❤️
A diagram should open the door to knowledge, not close it.
We believe accessible education should offer more than a description of what someone else can see. It should give students the opportunity to explore, question, and understand the same ideas.
Mimo doesn't just describe what's in a diagram. It helps uncover what's missing from the learning.
Every diagram has a story. Every student deserves to understand it.
Built With
- accessibility
- ai-agents
- ai-evaluation
- assistive-technology
- computer-vision
- css
- edtech
- educational-ai
- gemini-api
- generative-ai
- google-gemini
- html
- human-centered-ai
- image-understanding
- javascript
- knowledge
- missing
- multimodal-ai
- natural-language-processing
- prompt-engineering
- python
- screen-readers
- text-to-speech
