Research Engineer / Scientist - Storage for LLM
This position is no longer accepting applications(closed Aug 17, 2026).
Summary
Design and build distributed caching systems for large language models, optimizing GPU-aware KV caches, replication, and low-latency access to improve LLM performance.
- Build custom GPU aware caching layers
- Design distributed KV cache system
- Evaluate caching algorithms
- Extend open source KV stores
- Implement cache consistency protocols
- Implement cache synchronization protocols
- Implement eviction strategies
- Implement memory aware sharding
- Implement replication strategies
- Integrate cache with batched decoding
- Integrate cache with model parallelism
- Integrate cache with token streaming pipelines
- Monitor system performance
- Optimize low-latency access and eviction policies
- Store and retrieve intermediate states for transformer LLMs
Perks/Benefits:
- Conference attendance
- Innovation-driven culture
- Open source contributions
- Research resources