BrainFlow

Inspiration Traditional learning platforms deliver the same static content to every learner regardless of their background, pace, or goals. We wanted to build something that feels more like a personal tutor — one that researches your goal, builds a custom curriculum, and adapts as you learn. We were also inspired by gamification mechanics from apps like Duolingo and visual novel companions to keep learners motivated.

What it does PathForge AI is an adaptive micro-learning platform that generates personalized learning roadmaps powered by AI. Users enter a learning goal (e.g., "Learn oil painting" or "Become a data analyst"), and the system researches the topic using Exa AI web search,generates a structured roadmap with modules and lessons,creates interactive slide-based micro-lessons with AI narration (ElevenLabs TTS),produces adaptive quizzes that adjust difficulty based on performance,tracks mastery, XP, streaks, and levels through a gamification system,includes an anime-style companion that reacts to your progress,and offers a social feed where users can share and clone each other's courses

How we built it Frontend: React 18 + TypeScript + Vite, styled with Tailwind CSS and shadcn/ui, animated with Framer Motion Backend: Lovable Cloud (Supabase) providing Postgres, Auth, Storage, and 15 Deno Edge Functions AI Pipeline: Exa AI for web research, MiniMax for lesson/quiz generation, ElevenLabs for text-to-speech narration, ZhiPu for video generation Architecture: Three React Contexts (Auth, UserProgress, AIRoadmap) manage state; content generation runs as a parallelized pipeline — research → roadmap → lessons → outcomes — all orchestrated through edge functions

Challenges we ran into

Content generation timing: Generating an entire curriculum (roadmap + multiple lessons + quizzes) takes time. We had to design a background generation system with progress indicators so users aren't staring at a blank screen.

Audio synchronization: Aligning TTS narration with slide transitions required pre-loading audio and disabling auto-advance until narration completes, plus handling browser autoplay restrictions. State persistence across navigation: The AI generation pipeline would break when users switched tabs because React hooks unmounted. We solved this by lifting generation state into a Context provider. Adaptive difficulty: Building a mastery tracking system that meaningfully adjusts quiz difficulty based on spaced repetition curves and per-topic performance scores. Accomplishments that we're proud of A fully end-to-end AI content pipeline: from a single text prompt to a complete course with narrated lessons, quizzes, and social sharing The companion system with affection levels (Stranger → Soulmate) that creates genuine emotional engagement with learning A "clone course" feature that lets users instantly adopt any public course, building a community-driven content library Smooth mobile-first UX with swipeable lessons and bottom navigation

What we learned

Orchestrating multiple AI services (Exa, MiniMax, ElevenLabs) in a single pipeline requires careful error handling and fallback strategies Gamification mechanics (XP, streaks, levels) are surprisingly effective at driving engagement even with simple implementations Row Level Security in Postgres is powerful but requires careful policy design, especially for shared/public content Building adaptive learning algorithms (spaced repetition, mastery decay) that feel fair and motivating is as much a UX challenge as a technical one

What's next for PathForge AI Video lessons: Full AI-generated video content using MiniMax and ZhiPu video models Collaborative learning: Real-time study rooms where users on the same course can learn together Mobile app: Native wrapper for push notifications (streak reminders, daily check-ins) Instructor mode: Let educators upload their own curriculum and have AI enhance it with quizzes, narration, and adaptive pacing Multi-language support: Generate courses in the learner's native language