All projectsProject03 / 07
Learning path

what does cosmos actually solve?

Static learning material fails to adapt to individual student knowledge gaps.

LLM-driven adaptive problem generation and automated hint synthesis. Higher student concept mastery rates.

a look inside

Session preview
OrchestrationSeveral agents, one curriculumRetrieval, problem generation and hint synthesis run as separate agents over a vector-indexed curriculum, so each can be tuned without touching the others.
Feature mapping
Agent infrastructure
LatencyStreaming, not spinningOptimistic UI covers LLM stream latency.
Tutor dashboard

the research behind the system

Cosmos was a capstone before it was a product, so the design decisions are documented rather than assumed — problem statement, literature review, methodology, and the infrastructure they argued for.

Problem statement — where static material fails an individual learner.
Objectives mapped to the solution each one demanded.
Literature review — what prior tutoring systems got right and where they stopped.
Methodology, part one — how adaptivity was defined and measured.
Methodology, part two — the evaluation loop.
The infrastructure the methodology called for.

Problem statement — where static material fails an individual learner.