العربيةStart a conversation

Work

Three example programs across manufacturing, hospitality, and homes. Each shows what was captured, how quality was checked, and the data card that ships with the delivery.

Program 01 Manufacturing · Three production sites, Greater Cairo

Assembly line pilot

An egocentric and fixed-camera program capturing assembly, inspection, and station handover across two shifts, structured for a robotics team building manipulation and task-understanding models.

140 hoursFootage
22Contributors
310Sessions
5 weeksDuration

What was captured

  • Head-mounted egocentric video of assembly sequences and tool use
  • Two fixed overhead views per station for spatial context
  • Spoken step narration in Egyptian Arabic with English translation
  • Shift, station, and product metadata per session

How quality was checked

  • Written acceptance criteria agreed before capture
  • Three review passes: contributor self-check, annotator peer review, program lead audit
  • Inter-annotator agreement measured on a 10% overlap set
  • Rejected sessions re-captured, never patched
Data cardships with every delivery
Dataset
ARM-ASM-001
Modality
Egocentric + fixed video, audio
Labels
Action segments, hand keypoints, object boxes, step transcripts
Consent
Written, per contributor, per session
Review
3 passes · agreement 0.91
Formats
MP4 + JSON, COCO, RLDS
Version
1.2 · changelog included
Program 02 Hospitality · Two properties, Dubai

Hotel service workflows

A speech and interaction program capturing front-desk, concierge, and room-service exchanges in three Arabic dialects and English, structured for an assistant model that must understand real guest requests.

90 hoursAudio
18Contributors
3 + EnglishDialects
4 weeksDuration

What was captured

  • Consented multi-party audio at front desk and concierge
  • Third-person video of service handovers and room preparation
  • Guest request intent and outcome annotation
  • Dialect, role, and time-of-day metadata

How quality was checked

  • Native-speaker transcription for each dialect
  • Intent labels validated by hospitality staff, not only annotators
  • Guest-identifying content removed at source before structuring
Data cardships with every delivery
Dataset
ARM-HSP-002
Modality
Multi-party audio, video, text
Labels
Transcripts, speaker roles, intent, outcome
Consent
Staff written, guests informed and opt-in
Review
2 passes + domain validation
Formats
WAV + JSONL, Parquet
Version
1.0
Program 03 Homes and daily living · 40 households, Cairo and Casablanca

Home routines

A consent-first household program capturing everyday tasks and object interactions for an assistive robotics team, with privacy protection designed before the first recording.

260 hoursFootage
40Households
31Task types
8 weeksDuration

What was captured

  • Egocentric video of cooking, cleaning, and tidying routines
  • Object, surface, and room layout annotation
  • Task boundaries and interruptions
  • Household context: size, layout type, accessibility needs

How quality was checked

  • Faces and documents blurred at capture, before upload
  • Contributors reviewed and could withdraw any session
  • Coverage tracked across task types to avoid over-representing easy routines
Data cardships with every delivery
Dataset
ARM-HOM-003
Modality
Egocentric video, depth where available
Labels
Task segments, object masks, room layout, interruptions
Consent
Household written, withdrawal honored
Review
3 passes · coverage audit
Formats
MP4 + JSON, WebDataset
Version
2.0

Programs are described at the level of detail we share publicly. Client names, model details, and full data cards are shared under NDA.

Four weeks, fixed scope, fixed price, and enough validated data to test against your own benchmarks.

Start a conversationpartnerships@ariamena.com
Preview · example content