Kawsay AI – Live Indigenous Language Agent

Inspiration

Indigenous languages are disappearing, and most AI systems ignore them. We were inspired to build technology that amplifies ancestral voices instead of accelerating their disappearance.

What it does

Kawsay AI is a real-time multimodal agent powered by Gemini Live API that enables live voice conversations, bidirectional translation, and culturally contextual responses in Indigenous languages.

How we built it

We used Gemini Live API for low-latency voice streaming and contextual reasoning:

𝑉 𝑜 𝑖 𝑐 𝑒 → 𝐺 𝑒 𝑚 𝑖 𝑛 𝑖

𝐿 𝑖 𝑣 𝑒 → 𝐶 𝑢 𝑙 𝑡 𝑢 𝑟 𝑎 𝑙

𝐶 𝑜 𝑛 𝑡 𝑒 𝑥 𝑡

𝐿 𝑎 𝑦 𝑒 𝑟 → 𝑉 𝑜 𝑖 𝑐 𝑒 Voice→Gemini Live→Cultural Context Layer→Voice

To support low-resource languages, we implemented adaptive prompt scaffolding, controlled vocabulary injection, and phonetic guidance mechanisms.

Challenges we ran into

Limited digital datasets

Phonetic variability in speech recognition

Cultural sensitivity and ethical data boundaries

We addressed these through structured prompting, contextual memory buffers, and non-persistent session design.

Accomplishments that we're proud of

Achieving real-time voice interaction in a low-resource language

Building a scalable architecture without large training datasets

Designing an ethically conscious AI system for cultural preservation

What we learned

Real-time multimodal AI increases engagement and accessibility

Low-resource innovation requires architectural creativity

Cultural preservation must be community-centered and ethically grounded

What's next for Kawsay AI – Live Indigenous Language Agent

Expand to additional Indigenous languages

Partner with educational institutions

Develop community-controlled linguistic contribution tools

Explore lightweight and offline-compatible versions

Kawsay AI aims to ensure ancestral voices are part of the future of AI.

Built With

• ai
• claude