Velvet GTM brain

Velvet sources and curates audiovisual datasets and develops interactive models for AI labs and consumer companies.

category

Conversational AI training data

market profile

  • text: Multimodal model builders need video data that captures realistic interaction and spatial understanding.
  • text: Video data sourcing competes on provenance, licensing and the suitability of clips for model training.

personas

  • title: Multimodal research lead
    pains:
    • Conversational models respond slowly or lack spatial understanding
    • Finding relevant, unique audiovisual data takes research effort
  • title: Training data acquisition lead
    pains:
    • Must vet video data for quality and compliance
    • Needs structured datasets matched to a model's specific training needs
  • title: Consumer AI product lead
    pains:
    • Interactive experiences can feel unnatural
    • Needs a model and data partner for audiovisual conversation

icp

  • text: Frontier AI labs training audiovisual conversation or world models; no employee-size threshold is published.
  • text: Consumer technology companies deploying interactive audiovisual AI experiences.

message pillars

  • text: Improve models for interactive conversation.
  • text: Find distinctive video data across everyday and specialized domains.
  • text: Vet data before delivery.
  • text: Deliver training-ready material to research teams.

objections

  • objection: Can we license this data for our intended model use?
    response: Ask for dataset-level consent, provenance and licensing terms before purchase; the site describes compliance checks but does not publish a license.
  • objection: Will the footage match our model's domain and modality?
    response: Provide the dataset description and request a representative sample against your specifications.
  • objection: How much data can you deliver and at what cost?
    response: Request a scoped meeting; public customer pricing and capacity are not stated.

competitors

  • name: Protege
    website: https://withprotege.ai/model-builders/video
    why: Sources and verifies audiovisual datasets for frontier model builders.
  • name: Appen
    website: https://www.appen.com/ai-data
    why: Offers multimodal and audiovisual training data and collection services.
  • name: Troveo
    website: https://www.troveo.ai/
    why: Sources and curates video and other real-world data for AI labs.

keywords

  • audiovisual AI training datasets
  • conversational video training data
  • world model video data
  • multimodal training data licensing
  • interactive AI datasets
  • spatial reasoning video data

voice

  • Research-led
  • Aspirational about interactive intelligence
  • Concise and technical

signals

  • text: A lab is training an audiovisual conversation or world model and needs fresh video.
  • text: A consumer AI team is trying to improve interaction quality or reduce conversational latency.

hiring

  • Research roles
  • Engineering roles
  • Operations roles

pricing

model: unknown

motion

sales_led

channels

  • website meeting requests
  • LinkedIn

confidence

medium

Founders

  • name: Lucas Mantovani
    title: Founder/CEO
    role: CEO
    bio: Co-founder, CEO of Velvet. Ex conversational data + youngest member @ Meta FAIR. CS honors @ UChicago. Brazilian olympiad medalist in Math + Physics.
  • name: Lucas Tucker
    title: Founder/CTO
    role: CTO
    bio: Co-founder, CTO of Velvet. Previously at Adobe. Math and Computer Science at UChicago.

Back to YC companies