Organization Information

Coverage

Description

Non-profit community of Burkinabe engineers, researchers, students, and professionals building open-source AI models for the Moore (Mossi) language. Motto: "Building AI for Burkina Faso, by Burkina Faso." Has produced 12+ models spanning ASR (Whisper fine-tuned), TTS (XTTS, VITS, ParlerTTS), MT (NLLB, Mistral fine-tuned for French-Moore), and text generation (Gemma). Also maintains public Moore ASR and MT leaderboards for benchmarking. Close relationship with GO AI Corp -- Alban Nyantudre is a member of both organizations. BurkimbIA operates as the open-source community counterpart: while GoAI keeps model weights private, BurkimbIA publishes everything openly (Apache 2.0). Their SaChi system is an end-to-end Moore pipeline: ASR -> MT -> TTS.

🦄 UNICEF Relevance

Open-source complement to GoAI (the existing UNICEF Burkina Faso partner). Same language (Moore), same country, but fully open models (Apache 2.0) vs GoAI's private weights. UNICEF could leverage BurkimbIA's open models for independent evaluation and community improvement. Their public ASR and MT leaderboards are valuable for benchmarking (Task 2). Personnel overlap with GoAI (Alban Nyantudre) means the two orgs share knowledge but take different openness approaches.

Key People

• SAWADOGO Salif - Lead developer
  HuggingFace: https://huggingface.co/sawadogosalif. Time series forecasting, interpretable ML.
• Alban NYANTUDRE - ML Engineer
  HuggingFace: https://huggingface.co/anyantudre. Also at GO AI Corp. Specializes in speech corpora and audio quality. 11 Moore speech datasets.
• David Christophe Pegdwende Oubda - Member
  
• KABORE Josias - Member
  
• Mahamadi NIKIEMA - Member
  
• KOLOGO Boinzemwinde Josias Yannick - Member
  

Projects

• SaChi (Moore ASR/MT/TTS Pipeline): End-to-end Moore language AI: BIA-WHISPER (ASR, 3 versions fine-tuned from Whisper Large v3 on ~100hrs Moore audio), BIA-NLLB / BIA-MISTRAL (French-Moore MT), BIA-XTTS / BIA-VITS / BIA-ParlerTTS (Moore TTS). 12 models total. All Apache 2.0.
  Mooré
• Moore ASR Leaderboard: Public benchmark comparing ASR models on Moore language using the asr-benchmark-public dataset. Important resource for collecting Moore ASR benchmarks.
  Mooré
• Moore MT Leaderboard: Public benchmark comparing French-Moore translation models using the mt-benchmark-public dataset (1,483 rows across 6 domains).
  Mooré
• Moore Speech Corpora: Consolidated Moore speech corpus (243K downloads). Includes Bible speech, folktales, proverbs, riddles. Individual datasets under Salif and Alban's personal accounts.
  Mooré

Partnerships

• GO AI Corp (personnel overlap)

Notes

BurkimbIA and GoAI are distinct but closely linked. Alban Nyantudre bridges both. Many of Alban's HuggingFace datasets are mirrored across both orgs. BurkimbIA's models are open while GoAI's are private -- this dynamic is important context for the Moore language landscape. The ASR leaderboard (spaces/burkimbia/leaderboard-asr) is useful for the benchmarking study.