Kashcool

Kashcool is an AI-driven educational platform built for students who need more than static lesson PDFs. It combines live video classes, retrieval-augmented tutoring grounded in curriculum, and AR-enhanced study material so concepts stick beyond a single session.

I worked on the backend architecture and performance layer: a Django and Django REST Framework stack on PostgreSQL, deployed on Amazon Web Services (AWS) with EC2, RDS, S3, and Route 53. The brief was not only features — APIs had to stay fast as content, users, and session history grew.

Performance work cut API response times by roughly 70% through query optimization, caching strategy, and infrastructure hardening. That kind of improvement matters in edtech because chatty tutoring and video-adjacent metadata cannot feel sluggish on student phones.

RAG tutoring retrieves from each learner's curriculum and prior sessions instead of answering like a generic chatbot. Grounding reduces hallucination risk for academic content and makes follow-up questions more useful across a course, not only within one chat bubble.

Live classes use low-latency video (WebRTC) with recording, playback, and interactive overlays. AR visualizations help STEM topics land on mobile devices when a 2D diagram is not enough. Adaptive learning paths use session analytics to decide what a student should see next.

Alongside architecture, the engagement included mentoring junior developers — code review habits, API design clarity, and how to ship safely on Amazon Web Services (AWS). The product outcome after the RAG tutoring rollout was about +40% student engagement, with a backend that remains explainable under growth.

For founders comparing Django versus other stacks for AI products, Kashcool is a concrete example of Python's strength next to vector retrieval and data workflows, while still exposing clean APIs to React clients. It is education software shaped by metrics, not slide-deck AI.