Inspiration
Every AI today makes you dependent. You ask, it answers. You stop thinking. I wanted to build the opposite — an AI that refuses to answer your questions directly, and instead guides you to discover the answer yourself. The human brain learns best through struggle, not spoon-feeding. NeuroMentor was born from that idea.
What it does
NeuroMentor is a brain-mapped AI learning agent with four cognitive regions:
- Socratic Brain — never gives direct answers, only asks guiding questions
- Memory Brain — remembers your name, goals, and learning history across sessions
- Logic Brain — decides what kind of help you need at each moment
- Scaffold Brain — tracks how many hints you needed and scores your independence
Users get an Independence Score that rises as they need less help over time. The platform includes XP, levels, streaks, badges, a knowledge graph, project workspace, and session history — making learning genuinely addictive. It also has a workspace you can create and the files will be tehre and agent will guide you each step od code process not helpy you fully but each syntax and concpet will be pratically studing by earing badges.
How we built it
- Google ADK (Agent Development Kit) with Gemini 2.5 Flash as the brain
- MongoDB Atlas as the memory layer via MCP server integration
- FastAPI + uvicorn as the backend server with CORS support
- React frontend with neural particle canvas, animated UI, and gamification
- Four specialized agent tools: remember_user, recall_user, log_hint, get_independence_score, log_question, save_project_note
Challenges we ran into
- ADK session management required careful isolation per chat session
- MongoDB Atlas SSL handshake errors on Windows required custom TLS configuration
- Making the Socratic rule reliable — Gemini would sometimes break character and give direct answers, requiring careful prompt engineering
- CORS between React (port 3000) and ADK server (port 8000) took significant debugging to resolve correctly
Accomplishments that we're proud of
- Built a genuinely novel concept — an AI that actively works to make itself unnecessary over time
- The Independence Score system is something no other learning platform has
- Full gamification with XP, streaks, badges, and reward popups built from scratch
- Live knowledge graph that grows as the user explores topics
What we learned
- Google ADK is powerful but requires careful session and tool management
- Prompt engineering for consistent Socratic behavior is harder than it looks
- Gamification mechanics must feel earned, not trivial, to be meaningful
- MongoDB Atlas MCP integration opens up persistent agent memory in elegant ways
What's next for NeuroMentor
- Tools and space model reinforcement implementation
- model training , data set space workspace
- Daily challenges with bonus XP to build learning habits in strict lock
- Adaptive difficulty based on answer quality
- Learning paths — structured curriculum like a real course with real badge with inspiration of leetcode and gemini v3 version
- Mobile app for learning on the go with limits of agent IF video doest work try vimeo link below for better experiences.
Log in or sign up for Devpost to join the conversation.