Verification
A Verified badge means specific checks passed, not that we formed a favourable impression. Verification requires identity confirmation, completion of the required assessments for the role track, role thresholds met on the relevant dimensions, and a human quality review. The badge cannot be issued while any of those is outstanding.
Badges carry the date and assessment version they were granted under. Evidence ages: a profile whose evidence falls outside the freshness window is marked as due for reassessment rather than quietly presented as current.
How we use AI
AI is used as an assistant throughout the product: structuring imported CVs, generating assessment item variants from approved blueprints, applying written rubrics to responses, transcribing audio, parsing employer requirements, and phrasing match explanations.
It is not used to make final decisions about a person. Scoring weights and pass thresholds live in versioned configuration, not in prompts. Model output is schema-validated before it is stored. Low-confidence and borderline results go to a human reviewer. Every evaluator run records the provider, model, prompt version and rubric version so a published score can be audited afterwards.
We do not infer personality, honesty or emotion from faces or voices; we do not rank accents; and we do not claim to detect reliably whether work was produced by AI. Talent disclose how AI was used in their work, and we assess the quality of the result.
Automated decisions and your rights
Where an automated evaluation materially affects verification status or access to opportunities, the person affected can see a plain-language explanation of what was assessed, what the result was based on, its limitations and its confidence.
Human review
Any score-affecting flag or result can be sent to a human reviewer. Borderline and low-confidence results are routed to a person automatically.
Appeal
An appeal creates a tracked case with a named reviewer. An override never deletes the original result: both are retained with the rationale and policy basis.
Correction
Inaccurate profile data can be corrected. Nothing extracted from a CV by AI is published until the talent has confirmed it.
Export and deletion
You can export your data, close your account and request deletion, subject to records we are legally required to retain.
Privacy and data handling
Data minimisation. We collect what is needed for verification, matching, work delivery and lawful analytics.
Visibility is explicit. Profile fields carry a visibility setting enforced on the server, not merely hidden in the interface. Public profiles return only allowlisted fields.
Private by default. CVs, source files, portfolio assets and intro videos are held in private storage and served through short-lived signed links. A private video cannot be reached by guessing a URL.
Contact details are protected. Employers cannot bulk-export personal contact information. Communication stays inside Taskopedia until the engagement policy permits otherwise.
Consent is versioned. Privacy, AI-processing and publication consents are stored with their version, purpose and timestamp, and can be withdrawn.
Sensitive attributes are excluded. We do not use health or disability status, religion, ethnicity or political views as inputs to ranking or scoring.
Security
Data is encrypted in transit and at rest, with secrets held in a managed secret store. Access to personal data is role-based and least-privilege, and sensitive views and exports are logged. Administrative overrides require an explicit reason and are written to an append-only audit log.
To report a security issue, email contact@taskopedia.com with enough detail to reproduce it. We will acknowledge and give you a point of contact.
Complaints
If you believe an assessment result, a verification decision or a moderation action was wrong, contact us and say what outcome you are seeking. Complaints about a score are routed to a reviewer who was not involved in the original decision.
Engagement disputes between an employer and a talent member are handled through the dispute process attached to the engagement, so the scope, deliverables and approvals already on record form the basis of the resolution.
Fairness and ongoing review
Assessment rubrics are written around observable, job-relevant criteria. Before any evaluator model or prompt reaches production it is measured against a gold set scored by human reviewers, and an upgrade that regresses beyond tolerance is blocked. Published scores are sampled for human audit, and we track how often appeals overturn decisions as a calibration signal.
This page is a statement of how the product is built and operated. It is not legal advice, and it does not replace the contractual terms attached to a specific engagement.