# Multimodal data curation services > Multilingual text, audio, image, and video data sourced, annotated, evaluated, and governed for reliable AI. Turn multimodal language data into an accountable product asset, with clear purpose, provenance, quality, and ownership. A useful dataset is not a pile of examples. It records explicit decisions about modality, population, coverage, rights, annotation, quality, and change. ## Services - [Data sourcing & governance](https://admas.net/capabilities/data-curation/sourcing-governance/index.md): Language-data specifications, acquisition strategies, rights, provenance, documentation, and stewardship controls. - [Annotation operations](https://admas.net/capabilities/data-curation/annotation-operations/index.md): Guidelines, workforce design, calibration, tooling, quality control, and adjudication for multilingual labeling. - [Data quality & evaluation](https://admas.net/capabilities/data-curation/data-quality-evaluation/index.md): Evidence about whether a multilingual dataset is representative, consistent, safe, and fit for its intended model task. - [Multimodal data pipelines](https://admas.net/capabilities/data-curation/multimodal-data-pipelines/index.md): Traceable pipelines for multilingual text, speech, image, video, annotation, transformation, and model-ready delivery. - [Benchmark dataset development](https://admas.net/capabilities/data-curation/benchmark-dataset-development/index.md): Representative multilingual benchmarks that connect model measurements to real tasks, populations, and release decisions. ## Signals that this work matters - Dataset volume is known but coverage and provenance are not - Annotation guidelines produce inconsistent human decisions - Quality scores do not predict model or product performance - New languages are added without comparable governance ## Signals field brief - [Multilingual Data Curation for Real-World Use](https://admas.net/signals/multilingual-data-curation-real-world-use/index.md): A dataset is not ready because it is large and labeled. It is ready when its coverage, rights, provenance, representation, annotation decisions, failure modes, and intended use are explicit enough to support a model decision. ## Outcomes - **Purpose-built coverage:** Data composition follows the user population and the product decision. - **Defensible quality:** Guidelines, calibration, audit, and adjudication make judgments consistent. - **Traceable stewardship:** Provenance, rights, versions, limits, and changes remain visible over time. ## Frequently asked questions ### What does multimodal data curation cover? Multilingual text, audio, image, and video data sourced, annotated, evaluated, and governed for reliable AI. Admas treats it as a connected practice spanning Data sourcing & governance, Annotation operations, Data quality & evaluation, Multimodal data pipelines, and Benchmark dataset development. A project can start with one focused service and expand only where the evidence shows a dependency. ### Who is multimodal data curation for? This work is usually shared by data, research, model, product, safety, legal, and operations teams that need representative, governed multilingual or multimodal datasets. The exact team depends on who owns the affected user journey, data, system, content, market decision, and release risk. ### When should a team start multimodal data curation work? Start before a launch is locked when possible. Common signals include dataset volume is known but coverage and provenance are not, annotation guidelines produce inconsistent human decisions, quality scores do not predict model or product performance, and new languages are added without comparable governance. A focused diagnostic can still help when the work has already become a recovery project. ### What inputs does a multimodal data curation engagement need? Useful starting inputs are the model or evaluation decision the data must support, target populations and modalities, sampling and rights constraints, label definitions and edge cases, security, privacy, retention, and and acceptance requirements. They do not need to be complete: unknowns should be recorded as assumptions, risks, or discovery questions rather than silently filled in. ### How does multimodal data curation connect to other localization and internationalization work? The practice rarely stands alone. Product architecture affects localization; data affects model behavior; language quality affects release decisions; and program design affects whether improvements persist. Admas maps those handoffs explicitly so each specialist can work from the same acceptance criteria. ### What should AI automate in multimodal data curation, and what should people own? Models can propose labels, find duplicates, prioritize uncertain items, and assist quality sampling. Human contributors and domain specialists are required to define categories, supply grounded judgments, resolve ambiguity, protect participants, and detect systematic model-shaped bias. ### How is quality measured in multimodal data curation? Use evidence tied to the intended decision, not one universal score. Typical measures include coverage and representativeness, label validity and consistency, agreement and adjudication patterns, rights and provenance completeness, privacy and safety controls, and downstream model or evaluation utility. Results should be segmented by language, market, content or task type, and risk so an average cannot hide a serious local failure. ### How is multimodal data curation priced? Pricing depends on collection or asset volume, language and domain scarcity, participant and specialist requirements, annotation complexity, adjudication and audit depth, rights, security, and and delivery constraints. A defensible estimate separates repeatable production units from discovery, engineering, review, management, pass-through costs, and contingency. Admas scopes the acceptance criteria and review path before treating a volume number as a quote. ## Start a project - [Build a multimodal data curation project brief](https://admas.net/start-a-project/index.md?focus=data-curation): Share the product or model, modalities, languages, timing, and current constraint.