For employers

Hire this AI Trainer

Sign in or create an account to invite AI Trainers to your job.

Invite to Job
Cyber Space Initiative SRL

Cyber Space Initiative SRL

Agency
ROMANIA flag
Ploiesti, Romania
$25.00/hrIntermediate1+GDPR

Key Skills

Software

AWS SageMakerAWS SageMaker
Google Cloud Vertex AIGoogle Cloud Vertex AI
LabelboxLabelbox
Label StudioLabel Studio

Top Subject Matter

No subject matter listed

Top Data Types

ImageImage
TextText
VideoVideo

Top Label Types

Classification
Entity Ner Classification
Evaluation Rating
Question Answering
Text Summarization

Company Overview

Cyber Space Initiative SRL (ROCSI.eu) is a Romania-based, EU VAT-registered boutique agency building AI-powered civic-tech and resilience solutions. Our mission is to help communities and organizations improve governance, connectivity, and societal resilience using AI and secure digital infrastructure. On OpenTrainAI, we provide high-quality data labeling and AI training support focused on computer vision workflows across image and video: object detection (bounding boxes), segmentation/polygons, video tracking, and content classification. Our experience includes UAV/drone and aerial imagery, infrastructure and asset monitoring, disaster response and safety monitoring, and media analysis for content integrity. We support structured QA and audits (gold sets, error analysis, guideline iteration) to keep datasets consistent and measurable. We work remotely from Ploiești, Romania, with founder-led delivery and the ability to scale via vetted collaborators for higher volumes. We prioritize clear guidelines, measurable quality, fast communication, and GDPR-aware handling of client data (least-privilege access, NDA-friendly workflows).

IntermediateEnglishRomanian

Security

Security Overview

We take confidentiality and privacy seriously and handle client data on a need-to-know basis. We work in a GDPR-aware way and can sign NDAs / DPAs where required. Access to project data is limited to the people actively working on the task (typically founder-led delivery). Accounts are protected with strong credentials and MFA where available, and we keep permissions tight and project-specific. We aim to minimize data exposure. We only use the data required to complete the labeling/evaluation task and follow the client’s rules for sensitive fields (masking, redaction, exclusion, or restricted views). Data is accessed and transferred over secure connections (HTTPS/TLS). When a client provides an approved tool or workspace, we do the work there and avoid copying datasets outside that environment. Work is done on secured devices with encryption, automatic updates, and endpoint protection. For remote work, we use trusted networks and can use a VPN if the client requests it. Quality controls are built into the workflow: clear guidelines, review/QA passes, spot checks, and escalation for ambiguous cases. If needed, we can support gold sets and basic audit trails inside the chosen labeling platform. Retention follows the client’s policy. At project end, we return outputs and delete local copies and credentials as agreed, and we can confirm deletion on request.

Security Credentials

GDPR

Labeling Experience

Labelbox

Annotation QA, Gold Set Creation & Dataset Audits

LabelboxImageClassificationEvaluation Rating
Performed QA on image/video annotation datasets by reviewing samples, validating guideline compliance, and resolving disagreements. Built small gold sets for calibration, tracked recurring error patterns (missed objects, class confusion, inconsistent boundaries, mask quality issues, ID switches), and proposed guideline updates to reduce ambiguity. Produced QA notes and summary findings to support iterative improvements and more reliable training datasets.

Performed QA on image/video annotation datasets by reviewing samples, validating guideline compliance, and resolving disagreements. Built small gold sets for calibration, tracked recurring error patterns (missed objects, class confusion, inconsistent boundaries, mask quality issues, ID switches), and proposed guideline updates to reduce ambiguity. Produced QA notes and summary findings to support iterative improvements and more reliable training datasets.

2024
Google Cloud Vertex AI

Visual Misinformation & Content Integrity Classification

Google Cloud Vertex AIImageClassificationEvaluation Rating
Classified visual media (images and short clips) using a defined taxonomy to support information integrity workflows (authentic vs misleading, manipulated, out-of-context, synthetic indicators). Added short notes for borderline cases and applied rating-style fields when needed (confidence, severity, manipulation strength). Ran periodic QA spot checks, tracked patterns of ambiguity, and updated guidelines to improve consistency across batches.

Classified visual media (images and short clips) using a defined taxonomy to support information integrity workflows (authentic vs misleading, manipulated, out-of-context, synthetic indicators). Added short notes for borderline cases and applied rating-style fields when needed (confidence, severity, manipulation strength). Ran periodic QA spot checks, tracked patterns of ambiguity, and updated guidelines to improve consistency across batches.

2024
Google Cloud Vertex AI

Aerial Image Segmentation & Polygon Labeling for Infrastructure & Terrain

Google Cloud Vertex AIImagePolygonSegmentation
Created pixel-accurate masks and polygons on aerial/drone imagery for infrastructure and terrain features relevant to monitoring and resilience use cases (e.g., buildings/roofs, roads/paths, vegetation boundaries, water/flooded areas, debris or hazard zones). Followed strict boundary rules (tight edges, consistent inclusion/exclusion), documented ambiguous cases (shadows, partial visibility, reflections), and applied a consistent class taxonomy across batches. Performed QA spot checks to verify mask quality and label consistency, tracked recurring issues (boundary drift, class confusion), and refined guidelines to improve repeatability and downstream model performance.

Created pixel-accurate masks and polygons on aerial/drone imagery for infrastructure and terrain features relevant to monitoring and resilience use cases (e.g., buildings/roofs, roads/paths, vegetation boundaries, water/flooded areas, debris or hazard zones). Followed strict boundary rules (tight edges, consistent inclusion/exclusion), documented ambiguous cases (shadows, partial visibility, reflections), and applied a consistent class taxonomy across batches. Performed QA spot checks to verify mask quality and label consistency, tracked recurring issues (boundary drift, class confusion), and refined guidelines to improve repeatability and downstream model performance.

2024
Label Studio

Video Object Tracking & Event Tagging for Monitoring

Label StudioVideoClassificationObject Detection
Labeled short video clips for monitoring and simulation workflows. Tracked moving objects across frames with consistent IDs (e.g., people, vehicles, points of interest) and tagged events and scene changes (appearance/disappearance, interactions, route deviations, congestion, hazards). Added clip-level classification for context (scene type, visibility/quality flags, camera motion). Handled edge cases such as occlusions, fast motion, low-light, motion blur, and cuts between viewpoints. Performed QA spot checks to catch ID switches, missed objects, and inconsistent boundaries, and refined labeling guidelines based on recurring issues.

Labeled short video clips for monitoring and simulation workflows. Tracked moving objects across frames with consistent IDs (e.g., people, vehicles, points of interest) and tagged events and scene changes (appearance/disappearance, interactions, route deviations, congestion, hazards). Added clip-level classification for context (scene type, visibility/quality flags, camera motion). Handled edge cases such as occlusions, fast motion, low-light, motion blur, and cuts between viewpoints. Performed QA spot checks to catch ID switches, missed objects, and inconsistent boundaries, and refined labeling guidelines based on recurring issues.

2024
Labelbox

Drone/UAV Image Annotation for Infrastructure & Emergency Monitoring

LabelboxImageBounding BoxClassification
Annotated aerial/drone imagery for R&D use cases in infrastructure monitoring and emergency response. Built an object-detection dataset using bounding boxes and clear class definitions (e.g., vehicles, people, buildings/rooftops, road obstacles/debris, signage/markers, points of interest). Added image-level classification tags where useful (scene type, visibility/quality flags, “hazard present”, “obstruction present”). Handled edge cases such as occlusions, small objects at distance, motion blur, shadows, and overlapping objects, documenting decisions in short notes for consistent interpretation. Performed spot-check QA and second-pass reviews on samples, tracked recurring errors (missed objects, class confusion, inconsistent box tightness), and refined guidelines to improve consistency across batches. Work was carried out in the chosen labeling platform with secure, project-specific access.

Annotated aerial/drone imagery for R&D use cases in infrastructure monitoring and emergency response. Built an object-detection dataset using bounding boxes and clear class definitions (e.g., vehicles, people, buildings/rooftops, road obstacles/debris, signage/markers, points of interest). Added image-level classification tags where useful (scene type, visibility/quality flags, “hazard present”, “obstruction present”). Handled edge cases such as occlusions, small objects at distance, motion blur, shadows, and overlapping objects, documenting decisions in short notes for consistent interpretation. Performed spot-check QA and second-pass reviews on samples, tracked recurring errors (missed objects, class confusion, inconsistent box tightness), and refined guidelines to improve consistency across batches. Work was carried out in the chosen labeling platform with secure, project-specific access.

2024