RepoTutor

Inspiration

Last week, I tried to go deep into the open-source codebase of a framework I wanted to master. While trying to understand the large, unfamiliar codebase, I realized that the hardest part wasn’t the syntax: it was forming a mental model of how everything fit together.

Files referenced other files, functions jumped across directories and documentation explained what existed (and it was not very beginner-friendly, mind you) but not how it actually worked together.

I understood this problem is even worse for absolute beginners. Most learning resources rely on toy projects, while real-world codebases are large, interconnected and absolutely intimidating. That gap between learning programming and understanding real code is where many developers get stuck.

RepoTutor was born from the idea that any codebase should be teachable, regardless of its size or complexity.

What it does

RepoTutor turns any public GitHub repository into a step-by-step personalised tutorial they can interact with through voice.

Users paste a repository link, and RepoTutor:

Allows user to choose between Beginner and Advanced explanations
Breaks the project into logical chapters
Generates architecture and flow diagrams
Explains each component in clear, structured language
Lets users interact with the codebase using voice

Instead of reading files in isolation, users learn how the system works as a whole.

Who can use it

Beginner developers trying to understand real-world projects
Students learning from open-source repositories
Hackathon teams onboarding quickly
Developers joining a new codebase
Educators teaching software architecture using real examples

How we built it

RepoTutor is powered by Gemini 3’s multimodal reasoning.

GitHub repositories (which may be LARGE in size) are analyzed to understand structure, dependencies, and execution flow
Gemini 3 generates:
- Multi-chapter explanations
- Beginner and advanced learning paths
- Code-aware diagrams (architecture, data flow, execution)
Voice-based interactions allow users to ask the codebase questions naturally
Diagrams and text explanations stay synchronized, helping users visually and conceptually

This approach allows RepoTutor to translate code across text, visuals, and speech, not just summarize it.

Why Gemini-3?

Advanced Multimodality: Processes and synthesizes information across text, images and code with high precision, allowing us to get information through the codebase.
Deep Reasoning & Thinking: Offers unprecedented depth in analysis, with specialized "Deep Think" modes for complex problem-solving. This helps us piece together the entire codebase's functions/directories/api calls,etc.
1M+ Context Window: Supports large context windows, such as 1,048,576 tokens in Gemini 3 Flash, allowing for analysis of long documents or codebases.

Challenges we ran into

Breaking large, unstructured repositories into clean, teachable chapters
Ensuring explanations were accurate while remaining beginner-friendly or tech-heavy based on the user's preference
Keeping diagrams and explanations contextually aligned
Passing off more information to the Voice AI assistant
Avoiding information overload while still explaining complex systems

Accomplishments that we're proud of

Successfully turning complex repositories into coherent learning journeys
Building a Beginner / Advanced toggle that adapts explanations instantly
Integrating voice-based “Ask the Codebase” interactions
Generating diagrams that reflect actual code relationships
Creating an experience that feels like a personal tutor for any repo

What we learned

Beginners don’t struggle with code: they struggle with context
Multimodal explanations dramatically improve understanding
Teaching software requires structure, pacing, and visualization (not just text)
AI is most powerful when it connects ideas across formats, not when it replaces reading