@monigarr
Monica PetersFounder▲6L1
Founder, Software Architect, MoniGarr.com
Oct 14 · Discussion in Introductions · Hogansburg

Sovereign AI for Kanien'kéha: Monica Peters of MoniGarr

I'm Monica Peters, founder and software architect of MoniGarr.com, an independent research and engineering practice in Hogansburg NY. I've built Kanien'kéha language software since the early 1990s, including Onkwehonwehneha AI (multi-modal AI), mobile apps and the Kakwitene VR language-learning game, which launched on Steam in 2019. Today I work on language AI for low-resource languages and on responsible AI governance.

Right now I'm focused on making existing models work for polysynthetic languages: morphology-aware tokenizers, speech tech (ASR and TTS), and evaluation that respects dialect variation and community authority over the data. My open work includes mini-indig-llm-kit, a reference architecture for low-resource language LLMs. I build with Ollama, vLLM, LangGraph, Langfuse, Supabase and ElevenLabs, alongside Claude Code and Cursor.

What I'd love from this community: an introduction to anyone funding or deploying AI for sovereignty-sensitive institutions, language, culture or feedback from people who have evaluated speech models on low-resource languages.

Introductions

Comments2

  • @david
    David Trachsel
    Head of Growth at SaasRise, the community for SaaS CEOs
    Oct 15

    Very cool work! Great stuff.

  • @ryanallis
    Ryan Allis
    Founder of Lounge.ai and SaasRise, helping founders reach $100M+ exits
    Oct 18

    Welcome to Lounge, Monica. Polysynthetic morphology breaks most BPE tokenizers, since one Kanien'kéha word can carry a whole English sentence. Have you found that morphology-aware tokenization helps more on ASR word error rate, or on generation quality, when data is this scarce?

Log in to join the thread.