Staff AI Engineer
- London, United Kingdom
نُشرت قبل 3 أيام
عن الوظيفة
UFORCE is a combat-systems integrator transforming Ukrainian battlefield experience into deployable defence technology for allied nations. Built by Ukrainian practitioners and global technology leaders, UFORCE brings together unmanned air, sea, and land systems, software, command-and-control, and operational expertise into one adaptive combat architecture.
Our mission is to make defence dramatically faster, more scalable, and more cost-effective than traditional military systems — helping free nations deter aggression by making defence 100× cheaper than offence. We are building a new category of defence company: combat-proven, open to allied integration, shaped by real-world frontline experience, and focused on protecting democratic societies at speed.
About the role
The Staff Engineer, AI / ML — Self-Serve Toolchain will build the end-to-end system that lets customers adapt UFORCE models on their own private data without exposing that data to us.
This role spans data processing, foundation-model-assisted labeling, human-in-the-loop QA, active learning, training, evaluation, and model promotion. You will turn an in-flight principal-led capability into a repeatable toolchain that non-expert customers can run safely on-site.
What you’ll do
Own the ML adaptation pipeline from raw customer data to trained, evaluated, deployable models.Build foundation-model-assisted labeling workflows using tools such as SAM-2, Grounding DINO, open-vocabulary models, LLM steering, and human review.
Design self-serve operator workflows using yes/no/maybe feedback and natural-language corrections.
Create versioned datasets with lineage, data cards, label provenance, class distributions, and known gaps.
Build QA methods that catch systematic pseudo-label errors, missing annotations, and long-tail data gaps.
Develop active-learning loops that prioritize the highest-value frames for limited operator review.
Build reproducible train/eval pipelines with experiment tracking, model packaging, and promotion gates.
Design evaluation around held-out anchor sets, leakage prevention, baselines, slice metrics, and automated approve/reject decisions.
Package the toolchain for on-prem, air-gapped, regulated, or customer-held environments.
Close the field-failure loop by feeding live failures back into the next tune cycle.
Lead and grow a small specialist team around curation, auto-labeling, deployment, and onboarding.
What success looks like
Customers can run a full adaptation cycle without engineer intervention.Customer data stays inside the customer boundary.
The system produces trusted datasets with clear provenance and quality signals.
Pseudo-label quality is measured and systematic errors are caught early.
The promotion gate can approve or reject models based on evidence, not intuition.
Evaluation is protected by anchor sets, leakage controls, slice metrics, and baseline comparisons.
Field failures become reproducible inputs to the next training cycle.
Synthetic data is used only when it proves value against real held-out data.
The toolchain becomes a repeatable capability supported by a small, ramped team.
Required Qualifications
5+ years building production ML, AI, or computer-vision systems.
Strong Python and PyTorch.
Experience owning ML data pipelines, training pipelines, or evaluation infrastructure.
Deep CV data experience: detection, segmentation, annotation taxonomies, dataset curation, and data quality.
Hands-on experience with model-in-the-loop or foundation-model-assisted labeling.
Familiarity with tools such as SAM-2, Grounding DINO, FiftyOne, and annotation platforms.
Strong understanding of evaluation: held-out sets, leakage prevention, baselines, slice metrics, and promotion gates.
Experience with QA sampling, label-noise analysis, missing annotations, IAA, or alternatives when only one operator is available.
Experience with active learning or other methods for prioritizing labeling effort.
Ability to reason about dataset economics: quality vs. quantity, long-tail coverage, and cost-per-useful-example.
Experience with dataset versioning, lineage, experiment tracking, model registries, or data cards.
Strong ownership, communication, and systems thinking.
Experience leading engineers as a tech lead, staff engineer, or small-team manager.
Nice to have
Synthetic data, sim2real, or domain randomization experience.EO / IR / LWIR, remote sensing, maritime imagery, or small-object detection experience.
Privacy-preserving ML, federated learning, on-prem, or air-gapped deployment experience.
Experience building self-serve ML platforms or tools for non-expert users.
Experience with lakeFS, DVC, MLflow, Weights & Biases, Kubernetes, Kubeflow, Flyte, Dagster, Airflow, or Argo.
LLM application, context engineering, structured output, or LLM evaluation experience.
Exposure to radar, AIS, EO/IR fusion, tracking, sensor fusion, robotics, autonomy, UxV, defence tech, C2/C4ISR, or tactical systems.
The nature of combat has changed.
Tomorrow’s battlefield success depends on autonomy, speed, and adaptability.
And UFORCE is ready.
Are You?