Encord · London
Sales & BusinessGTMFullTimeposted 10d ago
Encord is the universal data layer for AI that helps 300+ AI teams train and run models on the right data. Our platform indexes, curates, annotates, and evaluates data across the full AI lifecycle, from development through production.
Trusted by Woven by Toyota, AXA, UiPath, Zipline, and more. We're an ambitious team of 100+ working at the frontier of AI and have raised $60M in Series C funding from Wellington Management, CRV, Next47 and Y Combinator.
We're hiring a Quality Systems Lead to own how Encord measures the quality of the human data we deliver to frontier AI labs, physical AI companies and enterprise AI teams — the standard itself, the systems that evaluate against it, and the audit function that produces the ground truth behind both. Data quality is what our customers buy. As we scale across data types — image and video, document, medical, LLM evaluation, robot teleoperation, egocentric capture — quality coverage cannot scale linearly with headcount. So this role has two halves that make each other work. You will build automated evaluation: model-assisted and LLM-based screening, agreement analysis at scale, anomaly and drift detection across annotation output. And you will build and run a dedicated audit team of around ten specialists in India, whose judgements become the labelled ground truth that trains and calibrates that automated layer. As coverage automates, the audit team moves up to the cases models can't judge and to generating gold sets for each new data type we take on. It is an unusual combination — engineering and consistent QC operations in one person — and it is the combination the job needs. You will also have an advantage your counterparts elsewhere in the industry don't: Encord owns the platform this work runs on, so the measurement you build can become native capability in the product rather than internal tooling.
Build automated dataset quality evaluation and root-cause detection — model-assisted and LLM-as-judge screening, agreement analysis at scale, anomaly and drift detection across annotation output
Hire, train, calibrate and manage a dedicated audit team of around ten specialists based in our India operation, held to inter-rater agreement and catch rate rather than volume audited
Turn audit output into labelled ground truth that trains and validates the automated layer, and manage the ratio of automated to manual coverage deliberately over time
Own the quality standard for every data type we deliver — written rubrics with worked edge cases, golden sets, and acceptance criteria agreed with the customer, alongside the Special Projects lead, before the first batch ships
Build scoring systems that rank annotator and reviewer performance and feed routing, staffing and offboarding decisions
Set the pass thresholds that certification gates on, so nobody works a project queue without having demonstrated they meet the standard
Report quality KPIs to leadership, and into the reporting our Special Projects leads take to customers: accuracy against client spec, inter-annotator agreement, rework rate, cost of rework, and coverage
Work with Project Management on remediation — you produce the measurement and the diagnosis, production owns fixing the project, and the standard stays independent of the people being measured
Own the unit economics of quality: cost per audited unit, and the coverage you buy per pound spent
Partner with Product and Engineering to bring quality measurement into Encord platform as native capability
You build and you operate. You'll write the evaluation pipeline, and you'll also run the weekly calibration session with ten auditors in another country
Your instinct on a coverage problem is to automate it — you reach for a model, a heuristic or a better sampling design before you reach for more auditors
Statistically literate in a practical way: sampling design, agreement statistics, and the judgement to spot a metric being optimised against rather than met
You can hold a calibrated standard across a distributed team you don't sit with — you know that ten uncalibrated auditors produce ten standards
A strong writer. Much of this job is producing rubrics a distributed workforce can follow without you in the room
You hold a standard under commercial pressure, and you bring the evidence that makes it stick with a delivery team or a client
Systems thinker: as interested in why a failure recurs across projects as in this project's defect rate
4+ years owning both technical and operational outcomes in a data, AI or service delivery environment where quality was measured rather than asserted
Hands-on Python and SQL. You build the analysis and the tooling rather than specify it for someone else
Practical experience applying models to a quality or evaluation problem — LLM-as-judge, model-assisted QA, automated evaluation, anomaly detection or classifier-based screening
Sampling methodology and agreement statistics (Cohen's and Fleiss' kappa, F1 against ground truth) applied to real production data
Experience hiring, training and managing a team, ideally an audit, review or QA team, and ideally distributed
Track record of building a quality framework or function, including the reporting leadership and customers run on
Bonus: direct experience of annotation, evaluation or model-training workflows, and of what frontier AI labs accept as evidence of quality
Bonus: a STEM degree, or a background in data science or research engineering
Bonus: multilingual delivery and linguistic quality assessment
Competitive salary, commission, and equity in a high-growth startup
Strong in-person culture — most of the team works from our London office 4+ days/week
25 days annual leave + UK public holidays
Annual learning & development budget
Travel for customer visits, events, and conferences across the UK and Europe
Company lunches twice a week
Monthly socials & bi-annual team offsites