← Portfolio ↓ Download PDF

Yao Xiang

Summary

Computer Engineering undergraduate at NUS, specializing in Advanced Electronics & IoT with a Minor in AI, building across AI infrastructure, model inference, agent runtimes, and hardware/software systems. Experience spans TensorRT GPU inference optimization, autonomous agent infrastructure, CI/CD, robotics, and independent research into persistent autonomous agents.

Education

National University of Singapore — B.Eng. Computer Engineering, specializing in Advanced Electronics & IoT · Minor in AI Aug 2024 — Aug 2028 (Projected)
  • Coursework: Data Structures & Algorithms, Software Engineering, Machine Learning, Digital Design, ARM Assembly, Embedded Systems, and Information Security.

Experience

Robotics Software Intern — KABAM Robotics | Python, C++, ROS 2, Docker, Jetson Orin, TensorRT, SAM3, YOLOv8, FFmpeg, CAN bus, LiDAR, SSH, Apps Script May 2026 — Dec 2026
Matrix-4 security robot
  • Optimized real-time SAM3 vision-language inference on Jetson Orin AGX using TensorRT, integrating FFmpeg pipelines and YOLOv8 overlays for low-latency deployment.
  • Ran Matrix-4's ATP Singapore PoC solo on-site, coordinating remotely with engineers while executing missions and debugging LiDAR SLAM, Nav2 MPPI, docking, and MWAN3 networking.
  • Shipped two field-debugging tools: a desktop GUI for full ROS topic inspection plus camera/VA diagnostics, and a mobile mission-control web app (React, Tailscale) for robot control, emergency stop, map/LiDAR views, live status, and haptic feedback—greatly enhancing debugging and on-site diagnostic efficiency.
Teaching Assistant, CS2113 (Software Engineering & OOP) — NUS | Java, GitHub, CI/CD Jan 2026 — May 2026
  • Teach software engineering practices including OOP design, JUnit testing, Git workflows, code review, and CI/CD to undergraduate development teams.

Hackathons & Recognition

Singapore Defence Tech Hackathon — Finalist, Top 6 (LOTR) (2026) · Vibe For Good — Semi-Finalist (2026) [certificate] · DSTA BrainHack CODE_EXP — Open Finalist (2026) [certificate] · Tencent "Age Well" · Synapxe × IMDA AI Innovation Challenge · VibeWithSG · AI Engineer Hackathon · TinyFish SG · Hack4Good · HacX — CS7
Certifications: Google AI Professional Certificate (2026) · Deep Learning Engineering Specialization (2026)

Selected Projects

Aiko — Persistent OS Agent & Multi-Model Runtime | TypeScript, Node.js, SQLite, React Native, Tailscale, Ollama / OpenRouter / Kimi, Qwen3 TTS Apr 2026 — Present
Persistent autonomous-agent runtime across Linux, mobile, Telegram, and VRM interfaces · details · aiko-core · aiko-avatar · aiko-app
  • Designed a 5-tier memory architecture using Hindsight BM25 + cosine RRF over SQLite vectors, episodic memory, and a self-map; redesigned dynamic context assembly to cut token payload from 109KB to 34KB (-69%).
  • Built a multi-model runtime with automatic failover across local Ollama, OpenRouter, and Kimi, stall watchdogs, and local Qwen3-TTS GPU voice inference (RTF ~0.19).
  • Engineered self-backup, self-diagnosis, multi-surface IPC, and calendar-driven autonomous scheduling across the open-source aiko-core, aiko-avatar, and aiko-app stack.
Replay — Singapore Family Safety & Asymmetric AI Engine | React, Vite, Node.js, Vercel, Render, OpenAI, SQLite Aug 2026 — Present
Asymmetric multiplayer scenario game for Singapore family safety against online harms · details Live Demo
  • Engineered a deterministic authoritative multiplayer room engine with role-private state and asymmetric evidence views across browser clients.
  • Implemented source-locked AI narrative generator for dynamic scam/safety simulation and structured real-time AI post-game debriefing.
AnAn — Autonomous Senior Care AI Companion & Care Kernel | Python, FastAPI, React, WASM, MediaPipe, SEA-LION, Qwen3-TTS, MapLibre GL, OneMap SG, SQLite Aug 2026 — Present
Autonomous senior care companion engine treating time, silence, and state as core inputs ({time, silence, state}) · details Live Demo Video Demo Tencent Age Well Challenge
  • Architected 7-phase decision kernel (WAKE → THINK → REVALIDATE → GATE → ACT → RECEIPT → COMMIT) with 3-tier permission gating and stale-decision revalidation before action dispatch.
  • Built silence FSM escalation with walking family contact tree, on-device browser CV health screening (rPPG heart rate, pose mobility, FAST stroke facial symmetry), local Qwen3-TTS voice cloning, and deterministic code safety floors.
Agora — Real-Time Agent-Native Development Environment | TypeScript, VS Code Fork, Mattermost, Node.js, WebSockets, Docker, Bash Jun 2026 — Present
Agent-native development environment integrating Mattermost and VS Code forks · details Video Demo
  • Built custom orchestrator service resolving parallel AI agent git merge conflicts in real time, featuring 3D spatial presence rooms over WebSockets and single-script container deployment (`up.sh`).
Kampung Kaki — Civic Emergency Coordination & Role-Aware Map | Python, FastAPI, MQTT 5, Redis Streams, MapLibre GL, OneMap SG, React, Docker May 2026 — Jun 2026
Distributed emergency-coordination system · details Video Demo DSTA Brainhack Open Finalist
  • Delivered unified operational picture for citizens, responders, and ops teams via single canonical state over retained MQTT 5 QoS 1 messaging, Redis streams, FastAPI, and OneMap SG APIs.

Technical Skills

Languages:
Python, JavaScript/TypeScript, Java, C/C++, SQL, Verilog, ARM Assembly
AI / ML:
PyTorch, DINOv2, SAM3 VLM, Hugging Face, llama.cpp, OpenAI / OpenRouter, Ollama, Whisper, MediaPipe WASM, SEA-LION, MERaLiON, Qwen3-TTS
Robotics:
ROS 2 Jazzy, Nav2, MQTT, Jetson Orin AGX, CAN bus, SLAM, Frigate video analytics
Frameworks:
React, React Native (Expo), Vite, Electron, FastAPI, Next.js, Mermaid C4, Tailwind, OpenCV
Infrastructure:
Docker, Vercel Edge, Render, Tailscale (WireGuard), Git/GitHub Actions, SQLite (WAL + FTS5), Linux/Unix
Hardware:
Arduino, Verilog / Vivado (Basys 3 FPGA), Raspberry Pi