Recommendation System
iNews Recommendation Engine
Personalized news recommendations using behavior signals.
Backend Developer | 2022 - 2024
Backend systems behind national campaigns, contests, and recommendation engines: the problem, what I built, and how it held up.
Built the recommendation backend for iNews.id, ranking article feeds per reader from location and reading behavior, at 55M+ monthly impressions.
Recommendations sit in the hot path of every article page, so the whole thing had a budget of a few milliseconds. That budget decided the architecture.
Elasticsearch does the ranking, MongoDB holds the reader profiles, and Redis serves whatever is hot, so a request rarely has to compute anything from scratch.
It runs as Go microservices, which meant a slow component could be replaced without taking the article page down with it.
Key impact
- Replaced one-size-fits-all article lists with feeds ranked per reader, using location and reading behavior as the signals.
- Kept recommendations fast at 55M+ monthly impressions by serving hot data from Redis instead of recomputing it on every request.
- Gave older but still relevant articles a way back into circulation, rather than letting them disappear after their first day.
Technologies used
- Go
- MongoDB
- Elasticsearch
- Redis
- Microservices