Problem
A knowledge-sharing feed reads the same recent discussion graph over and over for every user who opens it. On MySQL that meant a 2s news feed — slow enough that people treated it as broken.
What I built
- Core backend architecture for an interactive knowledge-sharing platform
- Redis caching over frequent data retrieval, improving news feed response time from 2s to 200ms (10×)
- A collaborative recommendation engine suggesting relevant discussions to users, which is what turned a feed into an engagement loop
Decisions and trade-offs
- Cache the feed, not the write path. The feed is read-heavy and tolerant of being seconds stale; authoring a question or answer still goes to MySQL and invalidates rather than silently serves old content.
- Redis as an explicit layer, not an accident. A 10× win only means something if you can say when it stops holding — eviction, invalidation, and what the database costs when the cache is cold are the parts I had to reason about.