SkillSet.ai
AI-generated learning paths, AI feedback on submissions and embedding-based matching with peers learning the same skill.
Overview
SkillSet.ai generates a learning path for a skill, broken into modules, lessons and tasks. Learners get AI feedback on what they submit, are matched with peers learning the same thing and work together in real-time project rooms.
Problem
Learning a skill alone from scattered content is slow, and it is hard to find people at the same stage to build with.
What I built
- Built AI-generated learning paths: modules, then lessons, then tasks.
- Added AI feedback on submissions.
- Built embedding-based peer matching on Vectorize.
- Built content ingestion as Cloudflare Queues jobs that re-index Vectorize.
- Built real-time rooms on Durable Objects and WebSockets.
Engineering
Background re-indexing
New content is ingested through Cloudflare Queues jobs that re-index Vectorize, so search and matching stay current.
Stateful rooms at the edge
Each project room is a Durable Object holding its WebSocket connections and shared state.
Hono API on Cloudflare Workers with Workers AI, Vectorize, Durable Objects, Queues, D1, KV and R2.
Outcome
Live on Cloudflare Workers.
Next: Mindalike