[TYLER CHEN]
Software Engineer // Systems & Backend
// ABOUT
I specialize in vertical system architecture across multiple domains with special attention to software-hardware interfaces and LLM inference engines. Previously, I worked as a security researcher investigating AI agent communication protocols at [Huawei] and I developed embedded C++ AUTOSAR applications for [Ford]. I also previously worked as a backend developer at [Cortico] and [pixelpad.io].
I am currently investigating topics related to static binary translation, operating systems, and possible improvements to LLM inference performance anywhere from software to hardware. I can be reached at t47chen[at]uwaterloo[dot]ca for inquiry.
// Selected_Work
A transformer inference engine built from scratch in Python. Implements GQA attention, RoPE, BPE tokenization, and KV caching for Qwen2.5, reaching ≥ 0.999 cosine similarity with the Huggingface reference implementation.
RepoStoryA git project analysis tool that analyzes repositories and their changes over time, pulling iterative development timelines while maintaining performance. Winner @ Hack The North 2025.
CHIP-8 EmulatorA CHIP-8 emulator implementing all 35 opcodes with COSMAC/CHIP-48 compatibility modes, a multithreaded CPU/timer architecture, and a phase-accumulated PCM audio subsystem, cross-compiled to WebAssembly via Emscripten.
CanSatAn Arduino-based seismometer space probe collecting atmospheric telemetry and seismic data via GPS, radio, and gyroscope, transmitting to a ground station over long-range radio. Best Science Mission Award.
JanusA browser extension that prevents third-party LLMs from cross-contextualizing user data by injecting semantic noise, isolating conversation identities, and generating adversarial prompts to scramble leaked context.
tootA full-stack tutor-matching marketplace with JWT auth, real-time chat, geographic search, and booking conflict detection, built with Postgres, React, and Flask.