Jake Lulla

Computer engineering student at UIUC with a 3.82 GPA and James Scholar recognition. Experienced in on-device machine learning and low-level systems, including development of a CLIP/ArcFace inference pipeline on the Apple Neural Engine and a RISC-V OS kernel bootloader. Proficient in C/C++ and Python, with a focus on performance-critical code optimization.

• PhotoVault, Built a fully on-device ML pipeline (CLIP ViT-B/32, SCRFD, ArcFace) on the Apple Neural Engine with fpl6 models (98.6% LFW), a custom Swift CLIP tokenizer, and optimized tensor preprocessing., Implemented BLAS-accelerated DBSCAN face clustering (~1s for 30K faces), scaling background indexing to 100K+ photo libraries. Designed serverless shared albums on CloudKit with zone sharing, delta sync, idempotent uploads, invitations, and reactions without backend infrastructure. Shipped embedding-powered search, vector queries, auto-share suggestions, and AI music slideshows, backed by a 74-test suite, widget, and share extension.
• RISC-V OS, Built an OS kernel from bootloader to userspace, implementing a preemptive scheduler, synchronization primitives, and interrupt/PLIC handling with correct locking and race-free concurrency., Designed virtual memory, paging, swap, a heap allocator, VirtIO drivers, and layered devfs with two mountable filesystems (ngfs, tarfs). Implemented fork/exec, pipes, a userspace C library, and shell utilities; debugged scheduler races, deadlocks, and memory consistency issues under QEMU.