LearnForge — Learn It By Watching It Built

Inspiration

Most AI "learning" tools do one of two things: hand you a finished answer, or bury you in a slide-deck course you'll never finish. Neither one teaches the way a good mentor does — sitting next to you, building something real, explaining each decision as it happens. We wanted to know: what if AI could actually do that? Not generate an answer, but build a project with you, at your pace, narrating it like a person would. LearnForge started as an answer to that question.

What it does

You tell LearnForge what you want to learn and pick your level — Beginner, Intermediate, or Advanced. It plans a small, real project around that topic and breaks it into 4–7 ordered teaching steps. Each step gets genuinely correct, runnable code and a level-appropriate explanation — not boilerplate, not a toy snippet. Then it narrates the whole thing: AI voice, synced captions, code appearing on screen in step with the explanation, so watching the video is the lesson. The finished lesson streams from LearnForge's public library, where anyone can like it, bookmark it, or remix it into their own personalized version. You don't just consume a tutorial — you watch something get built, and then you can make it yours.

How we built it

LearnForge is a full-stack Next.js app (App Router, TypeScript) with route handlers doing double duty as the API layer — no separate backend service to sync. PostgreSQL via Neon, accessed through Prisma, holds users, lessons, and library metadata. Auth.js handles sign-in with both Google OAuth and email/password.

The core pipeline runs asynchronously so a learner's request never blocks on a slow generation job: when someone submits a topic and level, we enqueue it with BullMQ on Redis (Upstash), and a standalone Node worker process picks it up and runs the pipeline end to end. First, Gemini (gemini-2.5-flash) plans the project and generates the step-by-step breakdown as structured JSON — forcing structured output here is what lets us reliably turn a model response directly into ordered lesson steps instead of parsing free text. Each step's code and explanation come out of that same structured generation pass. From there, Gemini TTS (gemini-2.5-flash-preview-tts) narrates each step's explanation, and Remotion — React-based video compositing — renders the narration, synced captions, and per-step code into a single MP4, frame-accurate rather than stitched together after the fact. Finished lessons and their assets land in Backblaze B2 behind signed URLs, so we get private, S3-compatible storage without exposing raw bucket access.

The reason this stack matters for the pitch: every layer of "watch it get built" is generated, not templated — the project idea, the code, the voice, and the video are all produced fresh per request by the same async pipeline, which is what makes the public library's remix feature possible (remixing re-runs the pipeline with a modified prompt, it doesn't just fork static content).

Challenges we ran into

One of our biggest challenges was creating structured learning content instead of simple AI responses. We wanted every lesson to feel like part of a real course rather than a collection of disconnected answers.

Another challenge was designing a learning experience that works for different topics and skill levels while keeping the interface simple and easy to use

Accomplishments that we're proud of

We're proud of building an AI platform that transforms a single prompt into a complete learning experience.

LearnForge AI generates personalized learning paths, AI-powered video lessons, quizzes, assignments, and progress tracking in one place. We're also proud of creating a platform that helps learners save time while making education more accessible and engaging.

What we learned

Building LearnForge AI taught us that AI can do much more than answer questions. It can organize knowledge, personalize education, and guide learners step by step through an entire learning journey.

We also gained valuable experience working with Gemini, BackBlaze, and Google Cloud while building a scalable AI-powered educational platform.

What's next for LearnForge — Learn It By Watching It Built

Building LearnForge AI taught us that AI can do much more than answer questions. It can organize knowledge, personalize education, and guide learners step by step through an entire learning journey.

We also gained valuable experience working with Gemini, Firebase, and Google Cloud while building a scalable AI-powered educational platform.

Built With

• ai
• backblaze
• cloudinary
• gemini
• neon
• nextjs
• prisma
• remotion
• tailwindcss
• typescript