// ai full stack engineer @ spark ring · shenzhen

Systems that
learn, ships that last.

I'm Perry Tu. I build AI wearables end to end at Spark Ring — agentic pipelines, edge ASR, Flutter apps, and the hardware SDKs that tie them together.

lately building →

View selected work Get in touch

// experience

Where I've
put in the hours.

Currently shipping AI hardware at a startup in Shenzhen — the blink-of-an-eye pace between a working demo and a product on a real finger.

01 May 2026 — Present

AI Full Stack Engineer

Spark Ring Shenzhen, China · current
  • Develop core functionality for start-up AI wearable products — iOS and Android app backends, Flutter UI, and hardware SDK plugins.
  • Build an end-to-end data processing and decision-making pipeline centered on the Speak · Capture · Act agentic workflow.
  • Design lightweight algorithms for voice data processing, edge ASR, and harness agents to push system accuracy.
  • Collaborate with hardware and product teams to drive real-world deployment of AI capabilities in customer scenarios.
flutteredge asragentic ai ios / androidhardware sdkvoice
02 May 2025 — Apr 2026

AI Research Intern

Institute for Infocomm Research, A*STAR Singapore
  • Surveyed state-of-the-art time-series foundation models (TSFMs), providing critical insights that shaped the team's technical roadmap.
  • Ran the full ML pipeline solo — data cleaning, feature engineering, parallel training, and evaluation in PyTorch — against data sparsity and deployment constraints.
  • Hardened foundation models with multi-scale algorithms; authored technical reports and drafted research manuscripts.
  • Translated theoretical approaches into deployable forecasting prototypes for AI-for-Energy applications.
pytorchtsfmforecastingresearch
Day-ahead electricity price forecasting for volatile markets using foundation models with regularization strategy arXiv:2602.05430 · AI4TS Workshop @ AAAI'26 (Oral & Poster) · co-author
03 Oct 2023 — Feb 2024

Embedded IoT Development Intern

Synergy Lab, Ngee Ann Polytechnic Singapore
  • Built and deployed standalone ESP32 modules capturing 10+ parameters in real time — voltage, current, temperature, humidity — across three on-campus PV sites.
  • Piped telemetry over MQTT into MySQL and visualised plant performance on a live Grafana dashboard.
  • Proposed a Random Forest + XGBoost hybrid for PV fault detection, hitting 92%+ accuracy on historical faults.
  • Allocated 60kW PV-plant data to the database and shipped a Dash.js monitoring dashboard for solar plants and charging stations.
esp32mqttmysqlgrafanaxgboost

// selected work

Things I've built,
end to end.

Six projects I can walk you through — from solar-plant loggers to foundation-model benchmarks. Everything else lives in the archive.

// about

From solar plants
to model checkpoints.

I started in electrical engineering — my polytechnic final-year project wired ESP32 sensors across a working solar installation and turned the readings into performance analytics. Somewhere between the first SQL insert and the first chart, I started moving up the stack and never really stopped.

At A*STAR I researched time-series foundation models and carried an AAAI workshop paper from literature survey to forecasting prototypes. Now I'm an AI Full Stack Engineer at Spark Ring, putting AI on a smart ring: agentic pipelines that listen, decide, and act; voice processing on the edge; and the app backends and hardware SDKs that hold it all together. I care about things that run in production on real hardware — my own server hosts this site, the recommender, and a few too many containers.

0tsfms fine-tuned
0pv fault-detection accuracy
0images annotated for vlm

// contact

Let's build something.

Always up for talking AI wearables, agentic systems, or edge inference — whether it's a role, a collaboration, or a good excuse to tinker. The fastest way to reach me is LinkedIn.

[email protected] LinkedIn GitHub