README / README.md
harrywinner's picture
Correct founding history (founded 2022; open datasets since 2024)
f37f220 verified
|
Raw
History Blame Contribute Delete
1.85 kB
metadata
title: Flagship
emoji: πŸŽ™οΈ
colorFrom: yellow
colorTo: green
sdk: static
pinned: false

Flagship

Flagship is a Moroccan-registered company, founded in 2022. We have served as a research partner across a range of projects, and since 2024 we publish open datasets and models and produce scholarship on AI risks and alignment.

Our work centers on three threads:

  • Open data & models β€” building and releasing datasets and models for underserved languages and domains, so the public-good layer of AI stays open and reusable.
  • AI safety & alignment β€” research and writing on emerging risks, the safe and accountable deployment of AI systems, and alignment.
  • Applied language technology β€” turning open research into tools real communities can use.

🌍 Lingo / NativeAI β€” preserving Cameroon's oral-first languages

Our flagship applied project preserves and revitalizes Cameroon's largely oral, low-resource languages through open voice and text data and French-pivot machine translation. Live at lingo.cm.

Open models

  • French-pivot MarianMT covering 59 Cameroonian languages (119 translation directions), served as compressed int8 β€” see cameroon-int8 plus the original fp32 releases.

Open datasets

  • cameroon_bibles β€” verse-aligned scripture text across 59 Cameroonian languages.
  • ghomala-spoken-bible β€” ~20 hours of aligned spoken GhomΓ‘lΓ‘' (audio + trilingual text).

All released under CC-BY-4.0. Read the research log at lingo.cm/blog.


Open data, open models, open questions about how to build AI responsibly.