Hi, I'm Theo.

Computer science student at the University of Michigan.

LinkedIn GitHub Email

Education

Work Experience

Projects

Lock-free SPSC Queue

Engineered a header-only, wait-free Single-Producer Single-Consumer (SPSC) lock-free ring buffer in C++20, achieving ~400M concurrent ops/sec (~2.50 ns/op) and ~36 ns one-way inter-thread handoff latency.

Eliminated false sharing and cache-line bouncing by isolating producer/consumer atomic indices using std::hardware_destructive_interference_size (256-byte alignment on ARM64/Apple Silicon) with local cached index tracking.

Optimized hot-path execution to sub-nanosecond (~0.81 ns / 1.24B ops/s) by leveraging power-of-two bitwise index masking (idx & mask) and acquire-release memory order semantics.

Implemented custom allocator support, enabling zero-allocation operation across stack arenas, memory-mapped shared memory, and huge-page buffers.

Architected full C++20 lifetime and exception safety for move-only and non-trivial types with templated emplace forwarding, in-place destruction, and validation via GoogleTest and Clang ThreadSanitizer/AddressSanitizer (TSan/ASan).

View

Plane Spotter

Built and shipped an end-to-end, real-time aircraft classifier that runs fully on-device on iPhone, identifying 16 commercial jet families at 88% test accuracy (0.87 macro-F1) from a live camera feed.

Designed a two-stage vision pipeline (YOLOv8n detection → MobileNetV4 classification) in PyTorch/timm, using transfer learning with a linear-probe warmup followed by discriminative-LR fine-tuning.

Closed the training-to-deployment domain gap with a custom physics-motivated augmentation suite (haze, motion blur, digital-zoom softness, backlighting, perspective warp) and a deterministic "field-sim" evaluation protocol.

Deployed models to the Apple Neural Engine via Core ML and built the full native app in SwiftUI/AVFoundation/Vision, featuring a live detection overlay, multi-lens telephoto zoom (up to 15×) for distant planes, and top-3 confidence readouts.

Simple Autograder

Designed and implemented a fault-tolerant code execution pipeline in Go using Temporal for durable workflow orchestration, ensuring zero submission loss across worker crashes and infrastructure failures.

Built a producer-consumer streaming architecture using Temporal Workflow Updates as a FIFO buffer, enabling low-latency real-time result delivery over a server-streaming gRPC interface without polling overhead.

Implemented a subprocess execution engine with per-test timeouts (5s), peak memory tracking via syscall.Rusage, and 128MB memory enforcement; non-fatal test errors are isolated from workflow health.

Defined a typed protobuf API for code submission and result streaming, exposing SubmitCode (server-streaming), GetSubmissions, and GetSubmission RPCs with proper gRPC status codes and deadline propagation.

Applied differentiated retry policies across activity types: exponential backoff (1s→60s, up to 10 attempts) for database writes; zero retries for subprocess execution to surface errors immediately without masking flakiness.

Skills: Go, Temporal, Durable Execution, Postgres, OS Processes

View

Flappy Bird AI

Collaborated with the Reinforcement Learning Subteam to build a neuroevolutionary agent, achieving superhuman, indefinite performance in “Flappy Bird” with under 10 minutes of training.

Wheel of Decisions

A website I first created to be a fun way to decide who goes first at our daily scrum meetings at Cisco, but it can be used in a wide range of situations. My manager loved it.

Skills: HTML, CSS, JavaScript

View

Certifications