# Hypa Intelligence

- **Event:** [Meta Global AI Developer Hackathon](https://cerebralvalley.ai/e/meta-ai-developer-hackathon)
- **When:** Oct 26, 12:00 PM – Nov 4, 9:00 PM (EST)
- **Where:** Online
- **Team:** [Chinemerem Ibe-Ekeocha](https://cerebralvalley.ai/u/chris-ibe)
- **GitHub:** https://github.com/hypaai/hypa-meta
- **Demo video:** https://www.youtube.com/watch?v=ibOOlYkXpfw
- **Gallery:** https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery
- **Page:** https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/3347

Ask Hypa is an AI assistant that speaks underrepresented African languages. You ask in your own language, by voice or text, and it acts, then answers in text or reads the answer aloud. Example asks: "Reply to this in Igbo." "Explain this prescription to my mum in Hausa." "Price this customer's order." It's built for people who read poorly, small traders, and families separated by language. It covers Yoruba, Igbo, Hausa, Nigerian Pidgin, Swahili, Twi, Efik, Ibibio, Tiv, Amharic, Somali, Zulu, Xhosa, Kinyarwanda, Wolof, Kanuri and more (over 146 languages).

We already ran African-language translation, speech recognition and text-to-speech in production. For this hackathon, we connected them to Meta's Muse models as one tool-calling agent.

How we use Meta's models:

• Muse Spark as the agent: Muse Spark (muse-spark-1.3) is the brain of Ask Hypa. It plans each request and calls our 15 language tools, including translate_message, translate_document, translate_image, look_up_term, speech_to_text, text_to_speech, translate_and_speak, process payment (with user permission) and process_voice_message. It then streams the answer back live.

• Muse Spark for vision (Hypa Lens): point the camera at a sign, menu or prescription. Muse Spark reads the text, and the agent translates it in real time.

• Muse Voice Transcribe for speech: spoken input and voice notes are transcribed with muse-voice-transcribe-1.0. The agent can then translate the words and speak them back, for full voice-to-voice help.

• Muse Image for image localization: our localize-image package uses Muse Image (muse-image-1.0) to rewrite the text in a photo, like a road sign, in another language while keeping the rest of the picture the same.

• Llama roots: our own translation models (Hypa-Llama3.1-8b-SFT) are fine-tuned from Meta's Llama 3.1 on African-language data. They power fast translation, the dictionary and Word of the Day.

One toolset, two ways to use it:
• Real users get the Ask Hypa app (iOS, Android, web) and a native phone keyboard with an on-device model. It types African languages correctly, with tone marks, and brings the agent into WhatsApp, email or any app.
• Developers get the same tools as an MCP server, so anyone building with Muse Code (or Claude Desktop) can add African-language voice and translation to their own apps.

Built with: NestJS + Prisma (API), Expo / React Native (mobile, with native keyboard and camera modules), Vite + React (web), in a Turborepo monorepo.

More than 500 million people speak African languages. Ask Hypa makes them first-class in AI, so people can get things done in the language they think in.

## More from Meta Global AI Developer Hackathon

- [Sprout](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/3344)
- [Timo](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/3345)
- [Real2Simulation](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/3346)
- [almona.eg](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/3348)
- [Derm Forge](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/3349)
- [Farfaz Learning](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/3350)

---

Markdown version of https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/3347. Site index for agents: https://cerebralvalley.ai/llms.txt · full text: https://cerebralvalley.ai/llms-full.txt