← Back to Students
Zehava F.
Bio
Software developer focused on systems, performance, and ML infrastructure. Led the SwapKV component for vLLM, virtualizing the KV-cache across HBM↔DRAM (HBMDramManager, TensorPool, KV Connector v1) to unlock much longer contexts on existing GPUs. Experienced with Python, PyTorch, Linux, Docker, and C/C++; fast learner, strong ownership, and effective team player. Seeking a Backend/ML-Infra/Systems role where I can drive impact at the infrastructure layer.
Skills
Python
C++
CUDA
PyTorch
vLLM internals
CUDA/PCIe fundamentals
Linux
logging/observability.
Bootcamp Project
High-performance key-value store with memory swapping capabilities
Mentored by: Pliops
Embedded Systems Bootcamp 2025 (Embedded)
Responsibilities:
Performed an in-depth vLLM internals study: traced scheduler v1/SchedulerInterface, KVCacheManager & BlockSpaceManager, attention backends (Flash/MLA), and existing KV connectors; built small probes to profile KV lifecycles and HBM↔CPU memory paths, documenting findings (diagrams/notes) that guided the connector design.
Designed and implemented HBMDramManager and TensorPool (memmap//dev/shm-backed) to stage KV-cache blocks in CPU DRAM with clean save/load/free APIs.
Built the vLLM v1 KV connector (HbmDramConnector) extending KVConnectorBase_V1, including metadata, block protocol, and KVTransferConfig integration.
Created a runtime layer residency map (layer → GPU/CPU) and operations hbmtodram / dramtohbm / freefromdram / get_location.
Click to enlarge
Additional Projects
CForEveryone Compiler Development: Designed and implemented a new programming language extending C with modern, simplified syntax, including var declarations, foreach loops, boolean types, and improved memory-management features. Built a full compiler and preprocessor featuring lexical, syntactic, and semantic analysis, AST construction, and automatic translation into standard C code, while maintaining full compatibility with existing C compilers.
English Level
Fluent