jija
Get the app
Businesses / Data

Data, made to your spec.

Captured, labelled or judged by people whose level is earned on reviewed work, in the field and from home.

Start a pilotBrowse our data sets
Services

Available services

Everything here is screen work or recording work — done from a desk, a sofa or a kitchen table, by the same levelled people who go out on the map.

Image and video review

Pass, reject and reason-code against your rubric, at whatever depth the rubric requires.

Annotation

Bounding boxes, points, polygons and segmentation masks to your schema.

Content and policy judgment

Applying your guidelines, with reason codes and an escalation path for edge cases.

Model output evaluation

Preference ranking, pairwise comparison and rubric scoring on model responses.

Red-teaming and adversarial prompting

Structured attempts to break a model, logged with attack class and outcome.

Transcription and audio labelling

Speech, speaker turns, timestamps and acoustic conditions.

Voice and speech capture

Prompted or spontaneous recording, varied by room, accent and background noise.

Video capture to brief

Walkthroughs, egocentric task footage and scene recording shot to your spec.

Search relevance and ranking

Query-result judgments against a graded scale.

Translation and localisation QA

Fluency, accuracy and register checks on translated output.

Record and address verification

Desk checks against public and supplied records.

Document extraction

Structured fields pulled out of scans, forms and unstructured documents.

Data sets

Browse our data sets

Collected to a spec you set rather than scraped, with a fingerprint, time and place on every asset.

Why levelled

Why reviewers are levelled

Most vendors sell hours from an undifferentiated pool and check quality afterwards. Here the trust tier is set before the work starts and reported after it.

  1. One reputation graphA reviewer's level is earned across field and remote work together. It does not reset per project, and it does not start at zero every time you open a new batch.
  2. Trust tier is a parameterRequire reviewers at tier 20 or above and you get them. You are specifying demonstrated reliability, not volume.
  3. Field experience is a real signalSomeone who has shot a compliant capture on the street knows what a compliant capture looks like on a screen. That transfer is why both sides run on one ladder.
Quality

How quality is measured

  1. Gold-standard injectionAbout one item in ten is a known-answer check, mixed in so reviewers cannot tell which.
  2. Trust is agreement, never volumeA reviewer's tier moves on gold agreement. Throughput earns nothing on its own, which removes the incentive to click quickly.
  3. Thresholds and re-auditA reviewer who falls below 85% on the known-answer checks is paused, and their recent work is re-checked at our cost, not yours.
  4. Consensus where it mattersTwo independent reviews on flagged or high-value items, with disagreements going to our team.
  5. Assignment separationA reviewer never receives their own submissions, or work they could have claimed themselves.
  6. Behavioural checksAnswers given faster than a person could read the item are sent back to the queue.

Targets: known-answer agreement ≥ 92% · buyer dispute rate under 1%

Security

Security and data handling

Private storage

Media is never public. It is served through Jija on links that expire after five minutes, and never cached.

No download button

Media is shown in the browser with no download path in the interface.

US storage, fixed retention

Everything is stored in the US and kept for a fixed period set in policy.

NDAs on request

Reviewers on your batch can sign your NDA.

Start a pilot

Start a pilot

Work is quoted individually, against the task and the volume. Tell us the shape of it and you get a price and a lead time back.

Book a 20-minute call