LoveMySkool had a collaborative learning platform and a hard problem: students get stuck at 11 PM, and teachers sleep. Hiring round-the-clock tutors doesn't scale. A generic chatbot bolted onto the platform would answer questions — badly, generically, and with no memory of the student. What they needed was a tutoring agent: one that knows each student, adapts to how they learn, and knows when to stop helping.
What makes a tutor different from a chatbot
Three design decisions separated this agent from “ChatGPT in an iframe.”
1. Per-student memory. The agent keeps a learning profile for every student — which concepts they've mastered, where they repeatedly stumble, whether worked examples or guided questions help them more. A question about fractions gets a different answer for a student who struggles with division than for one racing ahead.
2. Curriculum grounding via RAG. Answers are grounded in the school's actual course material — retrieved at question time with LangChain over the platform's content — so the tutor teaches the syllabus, not the internet's version of it.
3. Socratic guardrails. The hardest engineering problem was making the agent less helpful. A tutor that hands over answers produces grades, not learning. The agent guides with questions and hints first, reveals full solutions only after genuine attempts, and flags students to teachers when a struggle pattern persists.
The teacher's view
Teachers get a dashboard the agent maintains automatically: who's active, which concepts the class is collectively stuck on, which students need human attention this week. The agent doesn't replace teaching — it gives teachers compressed, actionable awareness they never had time to assemble manually.
Results
94% student satisfaction, 117% growth in active students after launch, and a platform that serves over a thousand learners at any hour without adding staff. The full client story is in the LoveMySkool case study.
The pattern transfers to any domain where personalised, grounded, always-on guidance matters — onboarding, internal training, customer education. If that sounds like your product roadmap, the architecture is proven.