Computer Vision Engineer

About QSee.aiQSee.ai builds computer-vision solutions for quality control in apparel manufacturing. We develop Physical AI in the real operating environment of factories, not in isolation from it.Location and Engagement• Based in Ho Chi Minh City, Vietnam.• Full-time consultancy engagement with QSee.ai, a Hong Kong-registered company.• English is our working language for technical discussions, documentation, code reviews, and collaboration with international customers and partners.• Regular travel to apparel factories in Vietnam is part of the role for use-case discovery, image collection, technical validation, deployment, and troubleshooting.• This is not a fully remote or laboratory-only position. Factory-floor involvement is essential to how we build and validate our technology.The RoleWe are looking for a hands-on Computer Vision Engineer with strong first-hand experience developing, debugging, validating, and deploying image-based ML systems.Computer vision is the technical core of this role. You must be able to work directly with datasets, annotations, model behavior, preprocessing, inference, post-processing, landmarks, coordinate systems, confidence levels, and real-world failure cases.This is not a narrowly segmented position. Depending on what is blocking delivery, you may work on model development, inference, validation, Python backend services, APIs, testing, runtime optimization, deployment, image labeling, or mobile and operational-system integration.QSee.ai is an early-stage company. We define goals and expected outcomes, then expect you to take ownership. You will not be managed task by task or given detailed instructions for every technical decision. You must understand the objective, identify the blocker, determine an effective path, execute, and deliver a verified result.Flexible Startup ScopeYou may work on model architecture in the morning, correct image annotations in the afternoon, and investigate a factory deployment issue the next day. Some tasks require advanced computer-vision judgment. Others are operational, repetitive, or straightforward. Both are part of building a reliable product.You will:• Design, train, fine-tune, adapt, evaluate, and integrate computer-vision models.• Work with pose estimation, keypoints, landmarks, object detection, and potentially segmentation.• Inspect datasets, identify representation gaps, define annotation rules, and improve labeling consistency.• Personally label, review, or correct images when necessary to unblock development or validation.• Diagnose failures using images, annotations, predictions, logs, and factory evidence.• Distinguish between problems caused by data, annotation quality, model behavior, image capture, preprocessing, inference, post-processing, or application logic.• Improve image validation, resizing, normalization, coordinate conversion, confidence handling, and landmark post-processing.• Export and serve models using ONNX Runtime and, where relevant, OpenVINO.• Maintain and extend Python and FastAPI services supporting model inference and media workflows.• Keep model outputs, API schemas, metadata, mobile integrations, tests, and documentation aligned.• Improve reliability and performance across CPU and GPU environments.• Support Docker-based execution, Redis caching, MinIO or S3-compatible storage, and environment-based configuration.• Travel to factories to understand workflows, collect and validate data, test models, support deployments, and investigate real-world failures.• Translate factory-floor observations into practical improvements across data, models, image capture, inference, and application behavior.• Work across technical boundaries when necessary to get the product working.Required ExperienceComputer Vision and ML• Strong first-hand computer-vision engineering experience.• Practical experience training, fine-tuning, evaluating, or materially adapting image-based ML models.• Experience with pose estimation, keypoints, landmarks, object detection, segmentation, or a closely related task.• Strong Python and PyTorch skills, plus practical OpenCV experience.• Understanding of dataset construction, annotation quality, image diversity, augmentation, model evaluation, and failure analysis.• Ability to inspect model inputs, output tensors, confidence scores, and post-processing logic.• Experience validating models against real-world images rather than relying only on headline metrics.• Experience deploying or integrating a vision model into a production or production-like system.Software Engineering• Ability to write reliable and maintainable Python code.• Experience with FastAPI or a comparable Python API framework.• Practical understanding of REST APIs, schemas, validation, and backward-compatible contracts.• Experience with pytest or similar automated testing tools.• Familiarity with ONNX Runtime and model-export workflows.• Comfort with Git, Docker, Linux, environment configuration, and secure credential handling.• Ability to understand unfamiliar code and make focused changes without unnecessarily rewriting working systems.Independent Execution Is EssentialWe need someone who can work against goals, not someone who requires continuous assignment and supervision.You should be able to:• Turn a product objective into an actionable engineering plan.• Enter an existing codebase and understand how it works.• Identify the highest-priority blocker without waiting for step-by-step instructions.• Form and test hypotheses using evidence.• Make pragmatic decisions with incomplete information.• Communicate risks and blockers early, while proposing solutions.• Balance delivery speed with technical reliability.• Know when to fix, simplify, document, escalate, or leave working code alone.• Move comfortably between advanced engineering and hands-on operational work.• Stay accountable until the result is implemented and validated.QSee.ai provides business context, product goals, relevant code and data, and direct feedback. We will not break every goal into tickets, prescribe each implementation step, or continuously supervise execution.What Good Looks LikeYou can open an unfamiliar repository, run the system, understand the model and API boundaries, and begin contributing without extensive handholding.When a prediction is inaccurate, you do not immediately assume the model needs more training. You investigate image capture and orientation, dataset representation, annotation consistency, resizing and padding, color conversion, model inputs and outputs, thresholds, coordinate transformations, post-processing, API serialization, and environment differences.When given an objective, you determine the most effective path, execute it, and provide evidence that the solution works. You move quickly without confusing speed with careless engineering.Nice to Have• DETR-style models or Ultralytics YOLO.• PyTorch-to-ONNX export.• OpenVINO optimization and CPU inference tuning.• Custom keypoint or pose-estimation architectures.• Redis, MinIO, S3, or image-upload pipelines.• Android or mobile integration.• Edge devices or constrained computing environments.• Manufacturing, industrial inspection, quality control, robotics, or Physical AI.• Windows and Linux development.This Role Is Probably Not for You If• Your computer-vision experience is limited to consuming external AI APIs or wrapping pretrained models in endpoints.• Most of your ML work has remained in notebooks or academic experiments.• You require detailed specifications before starting work.• You expect a manager to define and monitor every task.• You avoid work outside a narrow technical boundary.• You recommend replacing existing systems before understanding them.• You cannot clearly separate your personal contribution from your team’s work.• You consider image labeling, dataset inspection, testing, or documentation beneath the role.• You want a stable, predictable task mix.How to ApplyGeneric applications without a specific, defensible example of first-hand computer-vision work will not be considered.Please describe one computer-vision system you personally worked on during the last 18 months, including:• The problem and model architecture.• Your personal responsibility.• How the dataset was created, annotated, or improved.• How you evaluated the model and investigated failures.• How the model was integrated or deployed.• One difficult technical problem you personally diagnosed and resolved.• Evidence, if available: GitHub/GitLab repository, pull request, technical report, model card, demo, or publication.Confidential professional work is acceptable, but you must be prepared to explain and defend your contribution during a live technical interview.Our StandardStrong first-hand computer-vision capability. High ownership. Fast, effective execution. Verified outcomes.