Junior Computer Vision / ML Engineer

🚀 Project Description
Facie is an early-stage AI product working with face image analysis.
We are looking for a strong Junior Computer Vision / ML Engineer to build a reliable mechanism for measuring facial geometry from photos.

The first challenge is straightforward to describe but non-trivial to solve: given a face photo, accurately measure numerical parameters of different facial features, such as eye dimensions, mouth dimensions, nose geometry, jaw proportions, distances between facial landmarks, and other geometric characteristics.

This is a greenfield CV problem.
You will not receive a predefined model or a detailed list of experiments to execute. We expect you to explore several possible approaches, quickly validate them, discard what does not work, and find a solution that is good enough to move into the product.

At phase 1 - our engineering philosophy is simple:
20% of effort that delivers 80% of the result is often better than spending weeks searching for theoretical perfection.

✅ You are:

Confident with Python
Understand core Machine Learning concepts
Have practical experience with Computer Vision OR can demonstrate serious interest in the field
Can work with existing open-source models and libraries
Comfortable reading unfamiliar code and quickly running existing projects
Understand basic model inference and API concepts
Able to turn an unclear problem into several testable hypotheses
Comfortable when the first approach does not work
Able to explain how you would validate whether an experiment actually works
Focused on finding a practical solution rather than endlessly optimizing one approach
Comfortable using Git and working with an engineering team
Commercial Computer Vision experience is not required.

If your primary background is general ML or software engineering, we would still like to talk if you can demonstrate genuine interest in CV through things such as:
university or online CV courses
pet projects
GitHub projects
Kaggle
research or thesis work
reproduced papers
personal experiments

➕ Will be a plus:

OpenCV
PyTorch or TensorFlow
MediaPipe or other face landmark solutions
Experience with face detection, landmarks or segmentation
Understanding of 2D / 3D geometry
Experience reading and implementing research papers
Docker
Model serving or inference pipelines
Experience exposing ML functionality through an API
Experience designing reproducible ML experiments
Experience with Cursor, Claude Code, Codex or similar AI development tools

♟️ What We Expect from You:

Experiment-first thinking - generate several possible approaches instead of betting everything on the first idea
80/20 mindset - optimize for time-to-market and find the simplest solution that provides enough accuracy for the product
Ownership - do not wait for someone to define every next experiment for you
Validation mindset - always ask how we know that a result is actually correct, especially when perfect ground truth does not exist
Fast iteration - prototype, measure, learn, discard weak approaches and move forward
Engineering judgment - understand when ML is needed and when deterministic geometry or conventional code is a better solution
AI adoption - actively use modern AI coding tools when they help you iterate faster
AI skepticism - understand and verify AI-generated code instead of blindly trusting it
Clear communication - explain what you tried, what failed, what worked and what you recommend trying next

🔑 Responsibilities:

Research different approaches for measuring facial geometry from images
Build quick prototypes and experiments
Evaluate existing Computer Vision models and open-source solutions
Compare multiple approaches using measurable criteria
Design practical validation methods when perfect ground truth is unavailable
Analyze failure cases and identify why particular approaches fail
Select solutions based on accuracy, complexity and time-to-market
Build a reproducible facial geometry measurement pipeline
Document experiments, conclusions and technical decisions
Prepare successful solutions so they can be integrated by Software Engineers
Ideally, help turn successful prototypes into an inference service or API
Proactively suggest the next experiments instead of waiting for detailed tasks

🤖 AI-assisted Engineering

We actively use modern AI development tools.
You are welcome to use Cursor, Claude Code, Codex or similar tools to write code, explore libraries, integrate models and accelerate experiments.
We do not care whether every line of code was written manually.
We do care whether you understand:
what you are testing
why you selected a particular approach
whether the result can be trusted
how you validated it
what the generated code actually does
what you would try next if the approach fails

🍩 What We Offer:

Part-time or Full-time depending on the candidate
Remote work
A real Computer Vision problem with direct impact on the product
Significant ownership of the CV direction
Freedom to experiment with different technologies and approaches
Access to paid AI development tools where needed
Direct communication with the founder and engineering team
Opportunity to build a Computer Vision system from the greenfield stage into production
Opportunity to grow from Strong Junior toward independently owning an ML / CV product area

🗣 Recruitment Process:

Intro Interview → Live Problem-Solving Session → Small Practical Experiment → Final Interview

💻 Live Problem-Solving Session (Optional)

This is not LeetCode and not a theoretical ML exam.
You will receive a previously unseen Computer Vision problem and will be asked to think aloud and visualize your approach using a whiteboard, Paint, FigJam or a similar tool.
We want to understand:
how you decompose an unclear problem
how many reasonable approaches you can generate
what you would test first
how you prioritize experiments
how you validate results
what you do when the first few approaches fail
when you decide that a solution is good enough to ship

🧪 Practical Experiment (Optional)
Selected candidates will receive a small Computer Vision experiment.
You can use:
AI coding tools
Google
documentation
open-source models
existing libraries
research papers
We are not testing your ability to code without assistance.

We are testing your ability to turn an ambiguous CV problem into a working and validated solution quickly.

The expected output is a small working prototype plus a short explanation of:
approaches considered
approach selected
why it was selected
how the result was validated
discovered limitations or failure cases
what you would try next with one additional day