Inspiration Most developers don't have a learning problem. They have a direction problem. I've spent hours on YouTube watching full tutorials when I only needed 3 minutes of one video. I've gotten generic advice that had nothing to do with my background. I've started courses, lost momentum, and quit. The content exists. The personalized guidance doesn't. I built LearnPath AI because every learner deserves a coach that actually knows them.
What It Does LearnPath AI reads your resume, asks 5 deep questions, and in 60 seconds diagnoses exactly what's blocking your learning. It builds a visual personalized roadmap, finds the exact YouTube timestamp where your answer lives, pulls resources from Coursera, GitHub, LinkedIn and Medium, and narrates your entire diagnosis out loud through an ElevenLabs AI voice coach. A persistent chat lets you ask your coach anything — with full context of your background and progress — anytime.
How We Built It
Frontend: React, Tailwind CSS, Framer Motion, React Flow Backend: Python FastAPI deployed on Google Cloud Run Database: MongoDB Atlas for user profiles, sessions, chat history AI: Google Gemini 2.0 Pro for deep diagnosis and roadmap generation, Gemini 2.0 Flash with Google Search Grounding for live resource discovery Voice: ElevenLabs API for personalized mentor narration Video: YouTube Data API v3 + youtube-transcript-api for exact timestamp extraction CI/CD: Google Cloud Build for automated deployment
Challenges We Ran Into The YouTube timestamp pipeline was the hardest technical problem. Pulling a raw transcript, sending it to Gemini, having it identify the exact segment relevant to a specific user's knowledge gap, and building a deep link to that second — getting that chain reliable and fast took significant iteration. Gemini's JSON responses also required careful prompt engineering to ensure structured output every single time without hallucination. Deploying a FastAPI backend with all dependencies to Cloud Run while managing environment variables and CORS added unexpected complexity as a solo builder.
Accomplishments That We're Proud Of The YouTube timestamp feature works. You describe your exact knowledge gap, and LearnPath finds you the precise moment in a real YouTube video where that concept is explained — not the video, the exact second. Watching ElevenLabs speak a fully personalized diagnosis that references your actual resume, your specific background, and your exact blocker — out loud, like a real mentor — was the moment this stopped feeling like a hackathon project and started feeling like a real product.
What We Learned Personalization is only as good as the data you collect upfront. The 9-question onboarding that extracts self-assessment scores, learning history, specific blockers, and goals is what makes the Gemini diagnosis genuinely useful rather than generic. We also learned that for a solo hackathon build, the demo path matters more than feature count. Three features working perfectly beats ten features working poorly.
What's Next for LearnPath.AI
Adaptive roadmap — if you score below 70% on a readiness check, the roadmap regenerates with easier resources automatically Behavioral learning — track which resources you actually click and adapt future recommendations accordingly Peer challenges — challenge a friend to the same learning path, compete on XP and completion speed Mobile app — voice-first learning coach you talk to on your commute Enterprise version — companies use LearnPath to onboard engineers onto new tech stacks faster with personalized skill gap analysis across entire teams
Log in or sign up for Devpost to join the conversation.