Peter Zhang.
a software dev.
I build full-stack systems, game engines, and the occasional desktop cat. Currently building restaurant tech at Owner.com and AI tooling at Finks.AI.
Hey! Fancy seeing you here (^_^)/
I'm a Computer Engineering student at UWaterloo — software developer, game maker, tinkerer. Currently on Owner.com's Guest App Platform Team.
Down to chat tech, ski lines, or job opportunities. Especially the last one.





Places I've shipped
real things.
A few of the teams I've shipped real code with — from RTOS graphics drivers to AI-powered video pipelines. Open any role to see the highlights.
Hide3 highlights
- Shipped 4 location-confidence experiments across Online Ordering and the mobile app in 1 month; the winning variant increased order conversion by 4.6% and GMV per user by 3.2% across $84.3M in GPV.
- Redesigned the architecture of the mobile app's location-request system to prevent duplicate calls, added request-level observability, and introduced a feature-flagged hexagonal network layer with environment-specific test fixtures.
- Built an internal coding-agent workflow that helped deliver 46 mobile-performance tickets, covering investigation, implementation, CI repair, simulator validation, and human review.
Hide3 highlights
- Built and deployed Auxo on AWS, combining Claude, a Rust MCP backend, and a Next.js dashboard to generate 5 Instagram carousels per day and save about 3 hours of manual work.
- Co-developed and rolled out a shared OpenTelemetry SDK across Finks's Python services, replacing Logfire and standardizing error events through Amazon EventBridge.
- Sole engineer building a sales operations platform from scratch to replace 5 disconnected ad-sales tools, owning the architecture, data model, backend, frontend, and deployment for deal booking, advertiser approvals, click tracking, and performance reporting.
Hide3 highlights
- Authored the technical design and led a 3-person team that shipped AI Video Backgrounds in 5 weeks, from FFmpeg feasibility research through implementation and production rollout.
- Designed and benchmarked an AWS Lambda video-compositing workflow that placed background video behind transparent AI-presenter footage, testing resolutions, bitrates, and input sizes to establish production media limits.
- Implemented a Bun/TypeScript media-normalization pipeline that standardized inputs and bounded render times; partnered with Product and Design to launch configurable background controls.
Hide3 highlights
- Automated 500+ weekly codec regression tests across RTOS targets by engineering a Jenkins CI/CD pipeline.
- Eliminated configuration drift by containerizing test environments with Docker for reproducible testing setups.
- Validated 4 major OS releases by designing regression suites that caught critical graphics driver defects.
Hide2 highlights
- Designed a high-throughput "Batch Request" system in Java/TypeScript (MVC), implementing a caching layer to optimize cryptographic key generation.
- Worked with clients to modernize legacy security tooling by migrating an SVN codebase to Vaadin23, significantly improving maintainability and performance.
Things I built
on the weekends.
Charm
Self-hosted Discord assistant
Records voice calls, generates LLM meeting summaries, and semantically searches prior sessions. Async media pipeline with a GPU scheduler sharing VRAM between Whisper and local LLM jobs.
Storyline
Searchable long-form video archive
Local-first archive that links related recordings into searchable stories with an LLM. A CLI runs GPU analysis on desktop, stores artifacts on a Raspberry Pi, and resolves speaker identity across videos via diarization.
Echo
Discord transcription chatbot
GPU-aware semaphores arbitrating Whisper and Ollama. Event-driven job queue. Dependency-injected gateway for modular services management.
StudyGarden
Group study, in real time
10+ concurrent WS connections on Cloud Run, Redis state sync, TanStack Query + custom Notion API wrapper.

Desktop Pet
Jarvis, but a cat
A cat that wanders your desktop. Integrating Ollama and productivity apps so it's actually useful - when it's actually completed.

Statemachine Assistant
Visual state code editor
Node-based editor that generates ~40% templated of state code automatically.

Personal Website
v3, rebuilt from scratch
Next.js + TS + hand-rolled canvas visuals. You're looking at it. Source on GitHub if you want to poke around.

Raytracer Engine
C++ from scratch
Built a CPU raytracer in C++ following Ray Tracing in One Weekend, then bolted on multi-core processing to make it fly.

SoraGL
3D game engine, hand-rolled
Python + OpenGL game engine I built from scratch. Currently powering a rogue-lite procedurally generated shooter.
Daily Vibez
AI vlog from a Raspberry Pi
cuHacking submission. Pi 4B + Gemini API generates a daily vlog. Custom video-context cache, 300% search speedup.





