TimmyTutorials

Inspiration

TimmyTutorials was inspired by the Life 2035 theme, a future where learning is deeply conversational, personalized, and seamlessly accessible. I envisioned an AI tutor that feels like a friendly YouTube creator guiding you through a project, but with the added ability to pause, ask questions, and request modifications at any time. This idea grew from my own experiences wanting tutorials that adapt to me, not the other way around.

What it does

TimmyTutorials is an interactive coding tutor that builds full multi-file projects step-by-step while explaining each part in plain language. It supports voice narration, voice input, file uploads (PDFs, images, text), and a dynamic file explorer that updates as the project grows. Users can move forward, ask for clarifications, or request code modifications, and the system responds intelligently in the correct context.

How we built it

We built TimmyTutorials using a Flask backend, a structured LLM prompt engine, and a dynamic frontend that displays a live project tree. Each backend response is tagged with a clear intent, step, explain, or modify, so the UI can render actions predictably. Deepgram powers TTS and STT, turning explanations into natural voice narration and allowing users to talk back. A Hybrid OCR pipeline (Tesseract + Vision model fallback) extracts text from PDFs, screenshots, and scanned documents to seed the tutorial.

Challenges we ran into

One of the hardest challenges was evolving from a single-file tutorial to a fully multi-file project builder. In a single-file setup, the model can simply append code and continue the explanation, but multi-file projects require maintaining a consistent file hierarchy, tracking changes across many files, and ensuring the model always knows the current project structure. We also had to build a robust system for telling the model exactly which file to modify, when to create new files, and how to prevent it from mixing code between files. This meant upgrading prompts, adding strict response formats, and implementing a project-wide state manager so the AI could reliably build real applications that span multiple modules, folders, and templates.

Accomplishments that we're proud of

We’re proud of creating a fluid, conversational learning experience that feels closer to a personal mentor than a static tutorial. The multi-file project explorer, voice interaction, and structured step-by-step explanations make coding more approachable. We also built a robust OCR stack that successfully processes a wide variety of documents.

What we learned

We learned how essential response structure is when working with LLMs, predictability transforms the entire UX. We deepened our knowledge of multi-file context management, TTS/STT integration, and prompt engineering for educational interactions. We also discovered how to build reliable OCR pipelines by combining classical methods with modern vision models and how important clear backend to frontend contracts are for handling complex tutorial flows.

What's next for TimmyTutorials

Next, we plan to add real-time code execution previews, collaborative learning sessions, customizable tutor personalities, and support for more programming languages. We’re also exploring deeper voice interactivity, like hands-free navigation and natural back-and-forth conversations, as well as exporting finished tutorials into shareable project templates. The long-term goal is to make TimmyTutorials a fully adaptive AI teaching companion for the world of 2035 and beyond.