How to Write a Computer Vision Job Description
Computer vision candidates self-select on problem domain more than on company brand. A description that names the actual problem will outperform a polished but generic one every time.
Name the problem in the first paragraph
This is the single highest-leverage change available to you. Candidates in this field are choosing what to work on, and they filter fast. "Building perception for warehouse robots operating around people" tells a candidate more in ten words than three paragraphs about your culture. If you cannot describe the problem specifically, that itself is a signal worth acting on before you post.
Say which vertical you are in
Medical imaging, autonomous driving, industrial inspection and AR are close to separate job markets. Candidates have domain preferences and often domain experience that determines their fit. Burying the vertical costs you both relevance and search visibility.
Be honest about the research-to-application split
Overselling research freedom is the most common cause of early attrition in computer vision roles. If the job is fine-tuning existing models on your data — which describes most positions, and is perfectly good work — say so. You will lose some applicants at the top of the funnel and save yourself a resignation at month five.
Separate must-have from nice-to-have ruthlessly
Long mandatory requirement lists disproportionately deter exactly the candidates you want, and they are a well-documented source of demographic skew in applicant pools.
- Requirements should be three to five items. Everything else is nice-to-have.
- If C++ is genuinely optional, do not list it as required — it roughly halves your pool.
- Do not require a PhD unless the work is genuinely novel research.
- Never list a specific number of years alongside a framework. It signals nothing and screens out career changers.
Name the hardware and the data
For anything embedded, robotic or automotive, the platform is what candidates want to know. For modelling roles, the dataset is the draw. Access to a distinctive proprietary dataset is often a stronger recruiting argument than compensation, and it costs nothing to mention.
Get the title right
Title is how candidates find you, and inventive titles hurt. Use the term people search for, and add the qualifier that matters.
- Good: "Computer Vision Engineer (Medical Imaging)", "Perception Engineer — LiDAR", "Embedded Vision Engineer (Jetson)".
- Bad: "AI Ninja", "Vision Wizard", "Member of Technical Staff" with no further detail.
- Avoid "Machine Learning Engineer" if the job is really computer vision — you will attract the wrong pool and miss the right one.
Include salary
In several jurisdictions this is now a legal requirement, and everywhere it improves response rate and reduces wasted process on both sides. Candidates in a market with this much variance treat a missing range as a warning sign.
Common questions
How long should a computer vision job description be?
Shorter than most companies write. Aim for something readable in two minutes: what the problem is, what you will do, three to five real requirements, and the practical details. Length correlates with vagueness, not with rigour.
Should we list specific frameworks?
Name one or two that genuinely matter and treat the rest as transferable. Listing ten frameworks signals that you do not know what the job needs, and strong engineers read it that way.
How do we compete with better-funded companies?
On problem and data. You are unlikely to win on compensation against a well-funded lab, but a genuinely interesting problem, unusual data, real ownership and a shorter path to production are all things larger organisations struggle to offer.
Hiring right now?
Post your role to reach computer vision engineers directly, or start from a job description template.