Smart Arbitrage Technologies

Senior Computer Vision Engineer

About Us

We are a technology company with roots in algorithmic trading, where we built real-time systems that process massive, noisy data streams with zero tolerance for delay.
We are now applying this engineering DNA to a new MilTech division — an AI battlefield intelligence platform, from petabyte-scale video annotation to real-time inference on custom hardware at command posts.
We have exclusive data partnerships with frontline Ukrainian military units, an operational annotation pipeline with active learning and a trained detection model (v1).
The infrastructure is built and the data is flowing — we need experienced CV engineers to push the models forward.

The Role

We are looking for a Senior CV Engineer to join the ML team and focus on developing, training, and optimizing computer vision and vision-language models for battlefield target detection, classification, and contextual intelligence.

What You Will Do

Model Development & Training
- Train and iterate on detection, classification, tracking, and segmentation models on real combat footage
- Experiment with vision-language models (VLMs) for scene understanding and contextual reasoning
- Run distributed multi-GPU training experiments and manage the experiment lifecycle
- Improve model accuracy through architecture selection, hyperparameter tuning, data augmentation, and training strategy
- Contribute to the active learning loop: analyze confidence distributions, refine auto-labeling thresholds, identify hard cases for human review

Battlefield Contextual Intelligence
- Develop area-level temporal analysis: change detection across sequential overflights, environmental cue recognition
- Build sequence-aware model architectures (temporal attention, video transformers) for multi-frame reasoning
- Collaborate with military subject-matter experts to encode operator knowledge into model features and evaluation criteria

Model Optimization & Deployment
- Quantize and optimize models for deployment on custom AIPU inference hardware (INT8, mixed-precision, quantization-aware training)
- Validate that optimized models meet accuracy targets against gold-set benchmarks
- Support multi-stream inference validation at command post scale (100+ concurrent streams per rack)

Data & Annotation Quality
- Work with the annotation team to refine the labeling ontology and quality control criteria
- Analyze model errors to identify gaps in training data and drive targeted data collection
- Evaluate dataset diversity and coverage using embedding-based curation tools

What We Are Looking For

Must Have
- 4+ years in computer vision and deep learning, with production model deployments
- Strong PyTorch proficiency, including distributed training (DDP/FSDP)
- Hands-on experience with object detection pipelines (YOLO family, RT-DETR, or similar)
- Model optimization experience: quantization, pruning, or export to inference runtimes (ONNX, TensorRT, or similar)
- Solid understanding of training methodology: experiment design, ablation studies, metrics analysis
- Python as primary language; clean, production-grade code

Strong Plus
- Experience with vision-language models (CLIP, Florence, PaliGemma, or similar)
- Video understanding or temporal model architectures
- UAV / drone / robotics domain experience
- MilTech or defense-related work
- Change detection or remote sensing experience
- Experience with active learning or data-centric ML workflows
- C++ for performance-critical components

Tech Stack

- Training: PyTorch, distributed training (DDP/FSDP)
- Data processing: Ray Data
- Data curation: FiftyOne, embedding-based selection
- Annotation: CVAT with pre-annotation and active learning
- Experiment tracking: MLflow
- Inference: Custom AIPU hardware (model export via ONNX + quantization)
- Infrastructure: Kubernetes, on-prem GPU cluster (NVIDIA L40S)
- Storage: S3-compatible object storage (Ceph)

What We Offer

- A role focused on what you do best — building and improving CV models, not setting up infrastructure
- Real combat footage dataset — petabyte-scale, operationally relevant, impossible to replicate
- Dedicated on-prem GPU resources — no cloud queues or budget negotiations for compute
- Direct collaboration with military operators whose domain expertise shapes the models
- Competitive salary
- Stock options
 

About Us

We are a newly funded MilTech startup building an on-premises platform for automated annotation of petabytes of UAV video and training computer vision models for autonomous flight. We have a strategic partnership with the largest Ukrainian UAV manufacturer and access to what is the largest privately owned UAV video dataset in the world. We are at the very beginning — the technical plan exists, the funding is secured, the data is real — and now we need the right people to turn it into reality.

The Role

We are looking for a Data Annotator / CVAT Specialist to join the founding team. You will be the person who turns raw UAV video into the high-quality labeled datasets that our ML models are trained on. This is not bulk labeling work — you will help define annotation standards, manage the CVAT platform, build quality control workflows, and directly shape the data that drives our autonomous flight models.

You will work closely with the ML engineering team to establish the annotation pipeline from scratch: from ontology design and labeling policy through annotation execution, quality assurance, and feedback loops that continuously improve both label quality and annotation speed.

What You Will Own

CVAT Administration & Workflow
- Set up projects, tasks, and jobs; manage annotation workflows and task routing
- Configure AI-assisted annotation: integrate pre-trained models for auto-labeling and tracking
- Manage user roles, permissions, and review pipelines within CVAT
- Handle data import/export across formats (COCO, YOLO, MOT, CVAT XML)

Annotation Execution
- Annotate UAV video clips using bounding boxes, multi-object tracking (Track Mode with keyframes and interpolation), and classification labels
- Handle complex scenarios: occluded objects, crowded scenes, night/low-visibility footage, fast-moving targets
- Achieve and maintain high throughput without sacrificing label quality

Ontology & Labeling Policy
- Collaborate with the ML team to define and refine the annotation ontology (class definitions, examples, edge cases)
- Document labeling rules: occlusion/truncation handling, minimum box sizes, track start/end criteria, disappearance rules
- Maintain and update annotation guidelines as new classes or scenarios emerge

Quality Control
- Execute the QC pipeline: gold set calibration, double-labeling for inter-annotator agreement, review workflows
- Run automated validation checks: geometry (out-of-frame, zero-area, broken aspect ratios), temporal continuity (teleporting boxes, scale jumps, broken tracks)
- Track quality metrics: agreement scores (IoU + class match), rework rates, model-vs-human correction rates
- Identify systematic errors and feed them back into guideline updates and annotator calibration

Scaling & Active Learning Support
- Process pre-labeled data from ML models: review, correct, and approve auto-annotations based on confidence routing (high/mid/low)
- Prioritize annotation effort on hard cases (low confidence, rare classes, difficult conditions)
- Provide structured feedback to the ML team on where pre-labels consistently fail, driving retraining priorities

What We Are Looking For

Must Have
- 2+ years of hands-on experience in data annotation for computer vision (object detection, tracking, or segmentation)
- Strong proficiency with CVAT — project setup, Track Mode, interpolation, AI-assisted annotation, import/export
- Experience annotating video data (not just static images) — understanding of temporal consistency and tracking
- Familiarity with annotation formats: COCO, YOLO, MOT, Pascal VOC
- Attention to detail and ability to maintain high label quality at volume
- Experience with quality control processes: review workflows, inter-annotator agreement, gold sets
- Comfortable working in Linux environments and with self-hosted tools

Strong Plus
- Experience administering a self-hosted CVAT deployment (Docker-based setup, Nuclio for serverless auto-annotation)
- UAV / drone / aerial imagery annotation experience
- Familiarity with other annotation tools (Label Studio, Supervisely, V7, Roboflow)
- Understanding of ML model evaluation metrics (mAP, IoU, precision/recall) and how annotation quality affects them
- Experience with active learning or pre-annotation workflows
- Basic Python scripting (format conversion, data validation, automation)
- MilTech or defense-related annotation work

What We Offer

- Join an early-stage team where your work directly determines the quality of the data behind our ML models
- Define annotation standards and QC processes from scratch — not just follow someone else's guidelines
- The largest privately owned UAV video dataset in the world — petabyte-scale, impossible to replicate
- Work closely with ML engineers and see how your annotations translate into model performance
- Opportunity to grow into annotation team lead as the team scales
- Open-source-first approach with no vendor lock-in
- Competitive salary