Skip to content
YL

Yusong Leng

I take AI from research to products that real businesses run on.

AI scalp scans in production
24,000+
first-attempt success, spinal needle placement
66% → 92%
countries with regulatory approval
5

About me

I'm originally from Harbin, in China's far northeast, but Singapore has been home for more than a decade. My wife and I both did our PhDs at NUS — mine in visual scene understanding — and we became PRs and moved into our first home here. We're settled for the long term.

Outside of work I'm a visual person. I love photography and video — I shoot with everything from a phone to a DSLR, a drone, even a Polaroid, and I cut a small vlog of every family trip, just to hold onto the moments.

I'm also a keen plant-grower, and that hobby escalated: my current side project, Orbot, is a biomimetic robot I'm building end to end that turns plants into interactive companions.

Yusong Leng

Top-2 AI solutions

TrichoTrack

Standardized AI scalp diagnosis, run as a business at production scale

2021–2024 · Alvisual

Business impact

scans across 6 outlets in 15 months
24,000+
infrastructure cost per scan
~S$0.87
per outlet per month, all-in
~S$232
  • Virtually zero downtime across 15 months of production.
  • Objective scalp measurement in seconds by non-expert staff — standardized diagnosis across every outlet.
  • Near-zero marginal cost per new outlet: SaaS operating leverage built in.

Problem

A hair-care chain needed to standardize scalp diagnosis across its outlets and reduce its dependence on scarce trichologists in order to scale. Objective scalp measurement — hair and follicle density — is impractical to do manually, so in practice it was either skipped or bottlenecked on a few experts.

Approach

  • Explainable hybrid AI: deep learning detects only what a human can verify with their own eyes — hair strands, follicles, dandruff, pimples — and a rule-based expert system turns those observations into the diagnosis. Trust is what drives adoption: staff and customers can always see why.
  • A genuinely hard data regime — cold start, out-of-domain microscope imagery, severe class imbalance, no gold-standard labels — solved with concept decomposition for annotation, self-supervised pretraining, and focal loss.

Tech

Multi-tenant SaaS (DDD, FastAPI, PostgreSQL), PyTorch → ONNX CPU inference, Stripe & Twilio, Docker + CI/CD

My role

CEO & co-founder — took it from the first business conversation to production and 15 months of operations, leading a 5-person team.

uSINE

A GPS for the spine

2018–present · HiCura Medical

Business impact

first-attempt success (palpation → uSINE)
66% → 92%
countries with regulatory approval
5
leading Singapore hospitals in clinical use
5
  • First-attempt failures cut from ~1 in 3 to fewer than 1 in 10 — patients spared repeat needle passes.
  • The product I architected is now approved in 5 countries, in clinical use at 5 leading Singapore hospitals, with S$467k realized revenue.

Problem

Under manual palpation, roughly 1 in 3 first attempts at spinal needle placement fails — repeat punctures, complications, wasted clinical time. Ultrasound is the recommended remedy, but reading spinal ultrasound is genuinely difficult. uSINE is the AI guidance layer in between: it identifies spinal landmarks in real time and guides needle placement.

Approach

  • Prototype-to-production reinventions: replaced fragile sacrum-dependent panorama stitching with relational-state landmark tracking — drift-free, tolerant of free probe handling.
  • Replaced a hard-coded midline formula with symmetry-based detection over a much broader feature set — anatomy-agnostic, working at every spinal level.
  • Vectorized CPU backend for real-time performance on off-the-shelf laptops; production UI co-designed with clinicians and a professional UX team.

Tech

Real-time computer vision on off-the-shelf CPU laptops, ONNX, PyQt

My role

CTO & co-founder — led the full prototype-to-production rebuild across algorithms, backend, and UI.

AI + non-AI experience

  1. 2018–present(ongoing)

    HiCura Medical — CTO & Co-Founder, now Technical Advisor

    Architected uSINE, a real-time ultrasound spinal-navigation system, and led its prototype-to-production rebuild — now regulatory-approved in 5 countries. Co-founder as HiCura earned Forbes Asia "100 to Watch" and the University Startup World Cup title (2022).

    Medical AI · Computer vision · Regulatory

  2. 2021–2024

    Alvisual — CEO & Co-Founder

    Shipped TrichoTrack, a multi-tenant AI SaaS for scalp diagnostics — 24,000+ scans across 6 outlets over 15 months of production — leading a cross-functional team of five.

    AI SaaS · Product leadership · MLOps

  3. 2024–2025

    Anker — Senior Algorithm Manager

    Built Anker's Singapore AI team from scratch — hiring, mandate, and technical direction for a new regional computer-vision group.

    Team building · Computer vision

  4. 2024–present(ongoing)

    carbonZAP — Advisory Board Member

    Part-time, non-executive advisory — computer-vision roadmap for food-waste detection.

    Advisory · Computer vision

  5. 2025–present(ongoing)

    Independent R&D — Full-stack robotics build

    Building Orbot, a biomimetic robot, end to end — mechanical design, electronics, backend, AI, and app — while upskilling hands-on in LangChain and LangGraph.

    Robotics · GenAI · Full-stack

What drives me

The thread through every role I've held: apply AI to real business pain, where the investment case is clear and the benefit is measurable. A model isn't finished when it works on a test set — it's finished when the people it serves rely on it every day. That's why I care about the unglamorous parts: rollout, training the people who use the system, support, and the operating cost per unit of value.

I've learned to build solutions that outlive their first deployment — multi-tenant platforms, standardized and reusable components, systems designed so the next use case costs less than the last.

What I'm looking for now is depth: sustained, multi-year work with a strong team, hands-on growth in generative AI, and end-to-end ownership of outcomes — from the first conversation to the running system.

Ask me anything

This is a live GenAI build — a small language-model app built into this site, answering as me from a fact sheet I curated. Ask about my work, my projects, or what I'm looking for. Try German if you like.

ask Yusongwarming up

The assistant comes online here shortly.

AI answers from a fact sheet Yusong curated — verify anything important with the human.