Rutgers · Computer Engineering

Hi, I'm Aaron
(初荣恩).

I build things. Some of them run on a server, some of them run on a basketball court, and one of them runs on a pair of clippers. Based between the Bay Area and New Brunswick, NJ.

CE @ Rutgers / self-taught barber since 2023 / Hooper

Aaron Chu Bay Area → Rutgers

About

I'm from the Bay Area, studying Computer Engineering with a minor in Computer Science at Rutgers. I like building projects that help others out, not just ones that run on my machine. Outside of school I love to play basketball. I've played all my life on high school teams, intramural teams, and the Rutgers club team for a year. I also picked up barbering as a self-taught skill in 2023, and I've built a steady client base ever since.

Full Stack Development Embedded Systems Machine Learning Physical AI

Experience

  • Software Engineer Intern @ Magnefy Jun. 2026 – Aug. 2026
  • Intern @ Trainify Labs Inc Apr. 2026 – Jun. 2026
  • Undergraduate Research Assistant @ Rutgers Electrical and Computer Engineering Dec. 2025 – Present
  • Deputy Lead Systems Integration Engineer @ University Nanosatellite Program Sep. 2025 – Present
  • Drone Support Intern @ Integem Jun. 2025 – Aug. 2025

Projects

Tokeneyezed

Keeps a long-running coding agent on track toward a hard metric across compactions, crashes, and even a handoff to a different agent, while an observer blocks cheating and destructive steps and every attempt is stored in MongoDB Atlas so the plan can adapt when progress stalls. Built the memory and data layer end to end: embeddings, attempt ledger, bounded context builder, memory compactor, and retrieval eval. Built in a team of 4 for MongoDB's Harness Engineering & Model Wrangling hackathon.

Repository ↗
Control Loop goal loop, memory retrieval, and the observer's on-track/flag branch
flowchart TD
    F1["F1 — User submits goal (Dashboard)"]
    F1b["F1b — Scheduler heartbeat fires (async)"]
    F2["F2 — Frontend validates input"]
    F3{"F3 — Input valid?"}
    F4["F4 — Return validation error"]
    F5["F5 — Backend creates new session"]
    F6["F6 — Backend resumes session from checkpoint"]
    F7["F7 — Goal Loop Controller reads active goal"]
    F8{"F8 — Goal exists and incomplete?"}
    F9["F9 — Persist final state, return output"]
    F10["F10 — Retrieve memory (vector search)"]
    F11["F11 — Retrieve skills (vector search)"]
    F12["F12 — Check metrics history / budget"]
    F13{"F13 — Session budget exceeded?"}
    F14["F14 — Planner decides next action"]
    F15{"F15 — Tool required?"}
    F16["F16 — Treat response as step result"]
    F17["F17 — Tool Executor executes call"]
    F18["F18 — Observer evaluates call"]
    F19{"F19 — On track or flag?"}
    F20["F20 — Generate corrective note"]
    F21["F21 — Log event to sessions"]
    F22["F22 — Metric Scorer writes metrics"]
    F23{"F23 — Failure threshold exceeded?"}
    F24["F24 — Goal Loop Controller replans goals"]
    F25{"F25 — Goal completion criteria met?"}
    F26["F26 — Compactor summarizes + embeds to memory"]
    F27["F27 — Mark goal complete"]
    F28["F28 — [Stretch] Skill distillation to skills"]

    F1 --> F2
    F1b -.-> F6
    F2 --> F3
    F3 -->|No| F4
    F3 -->|Yes| F5
    F5 --> F7
    F6 --> F7
    F7 --> F8
    F8 -->|No| F9
    F8 -->|Yes| F10
    F10 --> F11
    F11 --> F12
    F12 --> F13
    F13 -->|Yes| F9
    F13 -->|No| F14
    F14 --> F15
    F15 -->|No| F16
    F15 -->|Yes| F17
    F16 --> F21
    F17 --> F18
    F18 --> F19
    F19 -->|Flag| F20
    F19 -->|On track| F21
    F20 --> F14
    F21 --> F22
    F22 --> F23
    F23 -->|Yes| F24
    F23 -->|No| F25
    F24 --> F7
    F25 -->|No| F26
    F25 -->|Yes| F27
    F26 --> F10
    F27 --> F28
    F28 --> F7
            

DeskPartner

An overhead camera and VLM planner drive a reBot B601-DM arm to autonomously sort a desk zone into trash, pen cup, and tray, running closed-loop until the desk is clean. Won Honorable Mention at the Embodied Metal Hackathon, hosted by Mission Robotics, New Theory, Savant, and North Star.

DeskPartner Demo ready, click a can or run the demo
grab berry berry grab cola cola grab grape grape
packed 0 / 3
shoulder −76.6° · elbow 96.6° · gripper open · idle

Knowsy

Interactive 5-stage RISC-V pipeline simulator for Knowsy, an open-source educational platform led by Professor Dov Kruger. Students type a program and watch RAW and control hazards, stall bubbles, forwarding paths, and branch mispredictions play out cycle by cycle, with processor-handled and compiler-handled modes. Led a team of 5 for development of the module.

GitHub org ↗
Pipeline Hazard Demo ready, pick a scenario or run the demo

cycle 0 / 10

VeriStudent

Marketplace for college students to safely buy and sell textbooks and find sublets, with .edu-locked sign-in. React/Expo frontend, Supabase Postgres backend with real-time chat.

View live ↗

PocketLab

Built with a team of 4, PocketLab turns a $10 ESP32 into a real-time IoT science lab: a React/TypeScript web app streams live sensor data over WebSockets into Recharts and Three.js visualizations. Includes guided, step-by-step experiment templates with noise filtering, offline fallback, and data export.

Zipply

Real-time chat app built on React and Firebase, with auth, instant messaging, and responsive design.

SuperHearo

Team of 4, Android app pairing voice control and speech recognition to help elderly and blind users with daily tasks.

Repository ↗

Basketball

A few highlights from the court.