MindMorph — Adaptive AI Learning Platform

A multi-agent learning platform where 11 specialised agents across 5 LangGraph workflows teach, plan, quiz, and grade — and a learner only advances when mastery is proven, not when a video ends.

01Why it exists

Online learning platforms optimise for completion, not competence. Content is static, sequencing is linear, and nothing checks whether you actually understood before moving on. MindMorph inverts that: the curriculum is generated per learner, and progression is gated on demonstrated mastery.

02How it works

1. Orchestrator routingA Groq-powered LLaMA 3.3-70B orchestrator classifies every learner query and routes it to one of 5 LangGraph graphs.
2. SCOUT — researchAcademic + Market + Practical + Factual agents research a topic in parallel → Consensus → Synthesizer → Reviewer → a Mermaid skill graph.
3. CONTENT & LEARNING PLANBoost / Builder / Sprint content modes; node progression through the skill graph is mastery-gated at ≥80% before anything unlocks.
4. LESSON deliveryStructured lessons grounded with live DuckDuckGo + job-scraper research, so material tracks what the market actually asks for.
5. EXERCISE — live gradingCoding challenges run in-browser via streamlit-ace with a live unit-test grader. Full LangSmith tracing over every agent hop; GitHub MCP integration.

03Results

• 1 orchestrator, 11 agents, 5 graphs — clean separation between routing, research, planning, teaching, and assessment
• Mastery gating (≥80%) makes progression evidence-based rather than time-based
• Live unit-test grader closes the loop: learn → attempt → graded feedback in one surface
• Full LangSmith observability across every workflow

04Stack