Hi, I'm

Hinny Tsang

Data Scientist / AI Engineer

I enjoy turning messy data into useful things - whether that's a trading signal, an internal tool, or a pipeline that just works. Background in physics, currently exploring the intersection of data science and quantitative finance.

See my journey

Educations

BSc in Physics

2015 - 2020

Hong Kong University of Science and Technology

Minor in Information Technology (core CS courses), Astronomy & Cosmology. University Scholarship recipient and Dean's List 2019 - 2020.

MPhil in Physics

2020 - 2025

The Chinese University of Hong Kong

Developed turbulence driving module in a computational magnetohydrodynamics code. Published in RASTI.

Visiting Scholar

Jul - Aug 2022

University of Virginia

NASA GPU Hackathon 2022: achieved 8x acceleration of hydrodynamic code with OpenACC GPU parallelization.

MSc in Financial Engineering

2025 - 2027

The WorldQuant University

Currently pursuing quantitative finance studies.

Fetching global weather...

Internship

Hong Kong Observatory

Jun - Dec 2019

Developed a flooding risk assessment pipeline for Hong Kong
Processed topographic data to improve storm surge modelling for flood-prone areas

server.ts

1import { Router } from "express";

3const app = new Router();

5app.get("/api/leases", async (req, res) => {

6 const leases = await db.leases.findAll();

7 res.json({ data: leases });

8});

10app.post("/api/leases", async (req, res) => {

11 const lease = await db.leases.create(req.body);

12 res.status(201).json(lease);

13});

15export default app;

Software Engineer

Oursky

Sep - Dec 2022

Built full-stack lease management system with React and Java Spring Boot
Set up Docker containerization and CI/CD with GitHub Actions on GCP Kubernetes

Analytics DashboardLIVE

Retention ↑

Churn ↓

Web Logs

0M+

Models

[INFO] retention_model — batch scored 24,519 users

Data Science Associate

SmarTone

Dec 2022 - Feb 2024

Built recommendation model improving customer retention by 5%
Developed X-learner uplift model with PySpark ETL reducing churn by 7%
Labelled 100M+ daily web logs using hierarchical clustering on Spark

BTC / USDT

Binance

OKX

1m candles

Data Scientist

Pollock Asset Management

Sep 2024 - Sep 2025

End-to-end quantitative models for statistical arbitrage (Sharpe 2.1)
Built ETL pipelines with Bloomberg SAPI/BQL
Deployed Apache Airflow orchestrating 30+ pipelines with MLflow tracking
Internal analytics tools used daily by PM

Qwen3.5-4B-GGUF-Q4_K_MSELF-HOSTED

A simple demo of self-hosted LLM inference, no complex architecture, just prompt engineering.
Based on Qwen/Qwen3.5-4B, quantized to Q4_K_M (2.7 GB) to run on a laptop with llama.cpp.
The model may occasionally be offline, hosting on a laptop means the fan gets loud!
Try it yourself: download the weights.

Ask me anything

Self-hosted via Tailscale FunnelQwen3.5-4B-GGUF-Q4_K_M

Founding AI Engineer

Stealth Startup

Nov 2025 - Present

Lead greenfield development of full-stack AI-powered MVP from scratch, architecting the whole scalable system