LearnForge — Learn by Building

Inspiration

Every tutorial has the same flaw: someone reads a script while a finished project sits on screen, already built. You watch, you nod, you feel like you understood it. Then you try to build it yourself and realize you never actually watched anything get built — you watched someone perform having already built it.

Real learning happens when you watch something come together, line by line, mistakes and all, at a pace that matches where you actually are. That's how a good mentor teaches. We wanted that experience available for anything you want to learn, on demand — so we built the thing that builds the video.

What it does

Tell LearnForge what you want to learn and your level — Beginner, Intermediate, or Advanced. It plans a real, working project scoped to teach that concept, and breaks it into 4–7 teaching-paced steps, the way a mentor sequences a lesson instead of dumping it all at once.

Each step gets real, runnable code and a level-appropriate explanation. That code and explanation are narrated with AI-generated voice and composited into a video — code appearing on screen exactly as it's taught, synced to narration, captioned throughout.

The lesson doesn't end at "watch and close the tab." Every video lands in a public library where it can be liked, bookmarked, commented on, and remixed — anyone can take your lesson and regenerate their own version, tuned to their level. Every lesson feeds the next.

How we built it

The hard problem isn't "call an LLM" — it's turning a vague learning request into something that behaves like a real curriculum. The planning step has to think in terms of a small, buildable project, sized to fit a handful of ordered steps a beginner can actually follow.

Every generated step is validated as real, runnable code before it's paired with an explanation — a lesson that teaches broken code isn't a lesson, it's a trap. From there, narration is generated, synced to the code appearing on screen, and composited with captions into the final video. The library, remix, and social layer turns each generated lesson into a starting point for the next one.

Built with: [ Nextjs, Gemini, Typescript, Tailwindcss, PostgrelSQL, Neon, Prisma, Remotion, vercel, github]

Challenges we ran into

Getting the AI to teach, not just explain, was the core fight. It's easy to get a model to output correct code and a paragraph about it — it's much harder to get it to sequence that code the way a good teacher does, holding back complexity until the right step instead of front-loading everything because it's "more efficient."

Keeping generated code runnable across wildly different requests was its own battle. And getting narration, on-screen code, and captions to feel synced instead of just technically aligned took real iteration — a video where the voice is half a beat off from what's on screen stops feeling like a mentor and starts feeling like a glitch.

Accomplishments that we're proud of

Type in something you want to learn, and minutes later there's a real video of it being built — at your level, narrated, captioned — that didn't exist anywhere until you asked for it. Getting that loop to feel like a mentor instead of a slideshow with a voiceover was harder than it looks, and it's what we're most proud of.

We're also proud of the library and remix layer: LearnForge isn't a generator you use once and forget — it's a growing place where one person's lesson becomes the starting point for someone else's.

What we learned

Generating correct code is the easy 80%. Teaching is the hard 20% — sequencing, pacing, knowing what to hold back until the learner is ready. We came away with a lot more respect for what good teachers actually do.

We also learned "personalized learning" means little if it's only personalized difficulty. What actually matters is personalizing the build — giving someone a project that's theirs, not a generic example reused for every learner at their level.

What's next for LearnForge AI — Learn It Built, Not Told

Pause-and-ask: let learners interrupt a lesson mid-video with a question, closing the gap between watching a mentor build and building alongside one. Beyond code: any step-by-step skill an AI can plan and narrate credibly, not just programming. Smarter remixing: regenerating someone else's lesson adapts their specific project to your level and goals, instead of starting from scratch. Path, not feed: library recommendations based on what you've already learned. Creator tools: let people guide the AI's teaching style, so LearnForge can carry a real mentor's voice, not just a generic one.

Built With

• backblaze
• gemini
• neon
• nextjs
• postgresql
• prisma
• remotion
• tailwindcss
• typescript
• vercel