> Markdown version of [/jobs/ext/4676418-applied-computer-vision-data-scientist](https://www.wearedevelopers.com/jobs/ext/4676418-applied-computer-vision-data-scientist). Every page supports `.md` or `Accept: text/markdown`. Links point to the HTML versions so they work for humans too. Agent guide: [/agents.md](https://www.wearedevelopers.com/agents.md). --- # Applied Computer Vision Data Scientist - **Company:** TEKSYSTEMS INC. - **Location:** Bentonville, AR, United States (Remote available) - **Salary:** $114,400 - $156,000 - **Contract:** Temporary contract - **Skills:** Data Analysis, Computer Vision, Virtual Camera Systems, Cloud Computing, Image Analysis, Data Deduplication, Data Files, Image Quality, Python (Programming Language), Machine Learning, Natural Language Processing, Object Detection, OpenCV, Tensorflow, Software Deployment, SQL Databases, Management of Software Versions, Image Acquisition, Pytorch, Large Language Models, Apache Spark, Model Validation, Data Lineage, Model Architecture, Machine Learning Operations, Software Version Control - **Published:** November 11, 2026 - **Apply:** https://www.careercircle.com/jobs/all/all/usa/ar/bentonville/95c34a46-5ab8-4010-a1a0-d12ea3599508 ## About the Role Scalability Forecasting Localization Data Science Data Analysis Visual Acuity Image Quality Failure Causes Experimentation Floor Scrubbers Post Processing Computer Vision Active Learning Machine Learning Business Valuation Course Evaluations Model Architecture Workflow Management Software Versioning Confidence Intervals, Hands-on experience building and evaluating models for at least one of the following: Object detection Image classification Product or visual recognition Object localization Visual search or matching Multi-camera perception Image-based inventory or location intelligence Candidates should be able to explain the model architecture, dataset, evaluation methodology, release path, production failure modes, and measured outcome. 2. Python and Modern ML Frameworks Strong Python development with hands-on experience using: PyTorch, Candidates must be able to move beyond notebooks and contribute maintainable, tested code supporting model training, evaluation, data analysis, and production validation. 3. Production Model Evaluation and Error Analysis Deep understanding of: Precision and recall False positives and false negatives Confidence thresholds Calibration Confusion matrices Error slicing Segment-level performance Ground-truth comparison End-to-end business accuracy Monarch specifically requires candidates who can distinguish model failures from image-quality, annotation, camera, catalog, pipeline, post-processing, and location-conversion failures. 4. Dataset Development and Annotation Quality Hands-on ownership of: Dataset creation and curation Training, validation, and test-set design Annotation guidelines Label-quality audits Hard-example identification Dataset versioning Deduplication Class or scenario imbalance Data lineage and reproducibility 5. Controlled Experimentation and Release Recommendations Experience designing experiments with: Clear hypotheses Controlled variables Frozen evaluation sets Acceptance thresholds Reproducible runs Champion-versus-candidate comparisons Production validation Evidence-based release recommendations Candidates Must Demonstrate Ownership Of Developing or materially improving a production computer-vision model Analyzing false-positive and false-negative patterns Creating or curating image datasets Improving annotation quality Selecting or recommending model operating thresholds Designing controlled experiments Producing model-release recommendations Validating models after deployment Explaining measurable improvements in model or business performance, python, data science, machine learning Top Skills Details python,data science,machine learning Additional Skills & Qualifications Secondary Skills and Nice-to-Haves Retail computer vision Product or SKU recognition Shelf intelligence Item-location systems Pallet recognition RF-DETR YOLO Detectron2 Vision Transformers OpenCV Active learning Hard-example mining Weak supervision Multi-camera systems Camera calibration Rolling-shutter image analysis Robot-captured imagery MLflow or comparable experiment tracking Distributed image processing SQL Spark or Ray Cloud-based model training Inter-rater agreement analysis Blinded evaluation design, Individual compensation offered for this position within this range will depend on many factors, including qualifications, skills, relevant experience, job knowledge, geographic location, internal equity, and other pertinent job-related factors. ## Description Next-generation item-location intelligence initiative evaluating how generated imagery can support accurate, scalable, production-ready inventory location capabilities. The program compares the existing Scrubber and IRAS production path with new processing approaches, using shared image inputs and verified physical ground truth to create a fair and auditable comparison. The Applied Computer Vision Data Scientist will own model quality, experimental rigor, dataset improvement, production error analysis, and model-release recommendations. This role will evaluate and improve computer-vision models that perform product recognition, detection, classification, localization, and related visual-perception functions using imagery captured by a five-camera system. This is not a generic data-science, LLM, NLP, forecasting, or recommendation-systems position. Candidates must demonstrate meaningful hands-on experience developing, evaluating, and improving computer-vision models used in production or production-equivalent environments., Develop, retrain, evaluate, and improve computer-vision models supporting Monarch and IRAS item-location workflows. Establish defensible model and end-to-end accuracy baselines across the Monarch evaluation arms. Analyze false positives, false negatives, recognition failures, and incorrect location results. Slice performance by club, camera role, product category, environmental condition, confidence level, and location-processing path. Evaluate exact physical-bay accuracy separately from adjusted one-bay-convention accuracy. Support the day-one physical census and daily disagreement, changed-result, and agreement-sample reviews. Design active-learning and hard-example workflows using prediction uncertainty, disagreements, production failures, and auditor feedback. Measure agreement between in-club auditors and remote image-review workflows. Determine whether errors originated in the model, image capture, rolling shutter, catalog data, annotation, pipeline, post-processing, or location-format bridging. Define model acceptance criteria for pilot and production deployment. Partner with the Pipeline Engineer to ensure inference outputs and metadata are correctly integrated. Partner with the MLOps Engineer to establish experiment lineage, model versioning, deployment controls, monitoring, and rollback criteria. Document dataset versions, model versions, experimental conditions, metrics, findings, and release recommendations. Honor the active-pilot change freeze and preserve the evidence necessary to reproduce evaluation results.