Redis
In-memory caching architectures, key eviction strategies, Pub/Sub event messaging, and distributed locking.
1 / Caching and Messaging
Redis served as both an in-memory caching layer and a message broker across several projects. In Foundry, Redis backed Celery task queues for async agent evaluation workflows. In Trajectory, Redis provided caching for frequently accessed application pipeline data.
2 / Caching Architecture
In Phoenix, Redis caching layers sat in front of the Hybrid RAG retrieval pipeline, caching embedding results and frequently repeated query-document similarity scores. Key eviction strategies (TTL-based expiration) ensured the cache reflected current document state without growing unboundedly.
3 / Building My Own Cache
Interestingly, building Shard and Cairn — distributed in-memory caches from scratch — gave me a much deeper understanding of how Redis works internally. Implementing Consistent Hashing, TTL expiration loops, LRU/LFU eviction policies, and Pub/Sub event messaging from the ground up made Redis's design decisions comprehensible rather than magical.
