Sam Merrick

I lead machine learning teams building computer vision that has to work in the field, not just on the benchmark.

At Simprints I lead the team training and evaluating biometric recognition models for global health, with field work in Ghana, Bangladesh, Ethiopia and South Sudan. Right now that means directing a $6m programme to build the first reliable contactless biometric for infants, running entirely on a low-spec Android phone.

Portrait of Sam Merrick
Face landmarks found by MediaPipe on this photo. Hover or tap to see the mesh. Landmarks: 0

How I lead ML teams

Problems without a playbook

There is no public dataset of infant biometrics, least of all from low- and middle-income countries, and an infant's features change as they grow. I break problems like this into experiments a small team can run, and keep everyone focused on the one number that matters.

Evaluation that reflects reality

I set up training, benchmarking and evaluation around data collected from the people a model will serve, tracked over time, so accuracy numbers still hold when the model reaches a clinic in rural Ghana or South Sudan.

Privacy as a design constraint

Biometrics for vulnerable people raises the bar on data protection. I've led template-protection research with the ICRC and shipped it as open-source code others can audit and reuse.

Built for the edge, not the cloud

Our models run fully offline on low-spec Android phones, using only the phone camera. I keep the team designing for that from day one, and bring a product background that asks whether a model ships, what it costs to run, and whether a frontline health worker can use it.

Selected work

Simprints

40% → 90%

true match rate at 1% false match rate, over 3 months of infant growth

4m+

biometric images collected

13,000

infants followed monthly for 12 months

Contactless biometrics for infants

Babies are the people most likely to be missed by health records, and the hardest to identify. I lead a $6m initiative to develop the first reliable contactless infant biometric recognition algorithm, from data collection through model training, evaluation and deployment.

Three things make it hard:

  • No data to start from. There are no public datasets of infant biometrics, especially from low- and middle-income countries.
  • The subject keeps changing. Infants grow fast, so a model has to recognise the same child months later.
  • Edge only. Capture and matching run fully contactless on a low-spec Android phone, with no cloud and only the phone camera.

Leading a team of two ML engineers, a backend engineer and, earlier in the programme, a data scientist, we built a longitudinal dataset of over 4 million images from a cohort of 13,000 infants in Ghana and Bangladesh, captured monthly for 12 months, and raised the true match rate at 1% false match rate from around 40% to around 90% over a three-month growth period.

Simprints, open source

20%+

lower deployment costs

SimFace: on-device face recognition

Launched SimFace, an open-source Android library for face detection, quality assessment, embedding and matching on low-cost mobile devices. It builds on the EdgeFace architecture and ships with SimQ, a configurable face-quality scorer, so poor captures are caught at the point of enrolment rather than in the data.

github.com/Simprints/Biometrics-SimFace

Simprints with the ICRC

1st

evaluation of PolyProtect for identification and fingerprints

Protecting biometric templates in humanitarian systems

Led a research collaboration with the International Committee of the Red Cross to choose and evaluate a biometric template protection scheme for mobile humanitarian deployments. We selected PolyProtect and tested it on real face data from a humanitarian project in Ethiopia, measuring accuracy, irreversibility and unlinkability, and published the first evaluation of PolyProtect for identification and for fingerprints.

Read the paper on arXiv · SimPolyProtect on GitHub

Experience

  1. 2023 – now

    Senior Product Manager, Simprints

    Non-profit deploying biometrics to verify global health outcomes. Lead a team of ML engineers, a backend engineer and a data scientist training, evaluating and shipping on-device biometric recognition models, with field work in Ghana, Bangladesh, Ethiopia and South Sudan. Infant biometrics, SimFace and template-protection research.

  2. 2022 – 2023

    Senior Product Manager, Sixteenth

    First in-house technical hire at an influencer management agency, reporting to the founder. Led the build of Spark Jam, an LLM-powered tool that generates content ideas for creators, taking it from concept to working product with rapid user testing. Prototyped a meditation app for a leading health and wellness creator, and managed a ~$100k technology budget.

  3. 2021 – 2022

    Product Manager, Student Beans

    Led the growth tribe on a B2B2C student loyalty platform, working with data science on personalisation. Rebuilt the website, cutting page-load times roughly threefold and lifting conversion ~50%; grew organic traffic ~350%.

  4. 2018 – 2021

    Product Manager, Virgin Red

    Founding team of the Virgin Group's loyalty platform. Launched Virgin Red to succeed Virgin Atlantic Flying Club (~2m members) and Virgin Red Vaults (~700k); introduced an ML-powered member-intelligence platform for reward targeting that contributed to a 30% rise in engagement. Helped grow product from 2 PMs to 6 squad-aligned PMs.

Publications and open source

Education and training

Get in touch

I'm based in Reading, UK, and work with partners and field teams across time zones. The best way to reach me is by email.