Image & Video
Annotation
Structured labels for visual datasets, with attention to object boundaries, class consistency, and continuity across frames.
- Bounding boxes & polygons
- Segmentation & keypoints
- Video tracking & frame review
HUMAN EXPERTISE. BETTER AI DATA.
Better data.
More reliable AI.
Specialist-led data annotation and AI evaluation for teams that need clarity, consistency, and quality at every step.
12+ years of professional experience. Direct founder involvement.
ANNOTATION WORKSPACEILLUSTRATIVE DEMO
Object detection Segmentation Keypoints
Quality is part of the workflow.Guidelines → Annotation → Review
BUILT AROUND YOUR DATA
Computer Vision / Image & Video / LLM Evaluation / Dataset QA
WHAT WE DO
The human expertise
behind better AI.
From the first annotation guideline to the final quality review, Altema helps turn complex data requirements into clearly defined work.
Image & Video
LLM
Evaluation
Human review of model responses against explicit criteria, with clear reasoning and consistent application of evaluation rubrics.
- Response & preference rating
- Factuality & instruction following
- Relevance & linguistic quality
Dataset Quality
Assurance
Focused checks that uncover annotation errors, ambiguous instructions, and inconsistencies before the next stage.
- Annotation audits & validation
- Error categorization & review
- Guideline consistency checks
Need a combination of services?Let’s scope your workflow
A CLOSER LOOK
Make quality tangible.
Explore simplified examples of the work. These original demonstrations illustrate the approach; they are not client projects or measured results.
SYNTHETIC SCENE / IMAGE & VIDEO ANNOTATION
Consistent class labels4 illustrative objects
IMAGE & VIDEO ANNOTATIONEvery boundary
Every boundary
has a purpose.
Useful labels start with clear rules: what to include, how to handle occlusion, and when to flag uncertainty.
- Define the taxonomy before labeling.
- Apply consistent inclusion rules.
- Review edge cases against the guidelines.
ILLUSTRATIVE EXAMPLE / RESPONSE REVIEW
USER REQUEST
“Summarize this passage using only the information provided.”
Reviewer note: flag claims that cannot be supported by the source passage.
LLM EVALUATIONClear criteria.
Clear criteria.
Reasoned judgments.
Evaluation is most useful when each judgment can be traced to an agreed rubric and the evidence available.
- Calibrate on sample responses.
- Separate factuality from writing style.
- Document ambiguity and review decisions.
ILLUSTRATIVE EXAMPLE / REVIEW LOG
ITEMOBSERVATIONACTION
Sample AClass mismatchRelabel
Sample BUnclear boundaryReview
Sample CGuideline ambiguityClarify
A useful QA report explains the issue, the applicable rule, and the next action.
DATASET QUALITY ASSURANCEFind the issue.
Find the issue.
Improve the process.
A review should reveal patterns as well as individual errors, so the findings can inform clearer instructions and targeted corrections.
- Agree on the review scope and sample.
- Categorize errors consistently.
- Report actionable findings.
HOW WE WORK
A clear path from
brief to delivery.
Start with a focused pilot. Align on requirements, resolve edge cases, and agree on quality criteria before expanding the scope.
Define
Discuss your data, goals, tools, and expected outputs.
Pilot
Work through a representative sample and test the instructions.
Calibrate
Review edge cases and refine the acceptance criteria together.
Deliver & review
Complete the agreed work with quality checks and documented findings.
Scope, timing, formats, and review arrangements are agreed for each project.Start the conversation ↗
Alexandru-Cătălin CiobanuFounder · AI Data Annotation & Evaluation Specialist
THE EXPERTISE BEHIND ALTEMA
Specialist-led.
Directly accountable.
A focused service, with a direct line to the person doing the work.
Altema is an independent AI data services business founded by Alexandru-Cătălin Ciobanu. His experience spans computer vision annotation, search relevance, language evaluation, and dataset quality review.
Projects are scoped with direct founder involvement. Capacity, timelines, and delivery arrangements are discussed openly for each engagement.
12+ yearsRelevant professional experience
Romania · RemoteAvailable for international collaboration
BEFORE WE BEGIN
Good questions.
Clear answers.
Every project starts with an honest conversation about fit.
Can we start with a small pilot?
Yes. A clearly scoped pilot is a practical way to review the guidelines, expected outputs, and working approach. Scope, pricing, and timing are agreed before work starts.
Which tools and output formats can you work with?
Relevant experience includes CVAT and Roboflow. Tell us about your existing workflow and required export format so compatibility can be assessed before committing to delivery.
How is quality assessed?
Quality criteria are defined for the task. Depending on the project, review may focus on class consistency, boundary rules, temporal continuity, or adherence to an evaluation rubric.
How should we share confidential information?
Keep your initial enquiry high level. NDA requirements and an appropriate transfer method can be discussed before any confidential material is shared.
Do you offer large-scale annotation teams?
Altema is a specialist-led independent business. Capacity and suitability are assessed for each project, with transparent delivery arrangements rather than an assumption of a large in-house workforce.
LET’S DEFINE THE NEXT STEP
Your data.
Your challenge.
A focused pilot.
Tell us what you are building and where human expertise could help. Start with a brief, not a dataset.
Primary business emailoffice@altema.ro
Secondary emailalexandrucatalin.ciobanu04@gmail.com
Phone+40 761 615 598
BASED IN ROMANIA · WORKING INTERNATIONALLY