Project thesis
AI agents preserve the archive. Communities decide what lives in it.
LangSafe LingHacks turns endangered-language data scattered across dictionaries, papers, videos, and oral-history archives into searchable vocabulary, grammar, source provenance, graph relationships, and teaching material.
Demo flow
- 1Open the Dashboard and choose Jejueo.
- 2Run the preservation demo and watch the agent feed fill in.
- 3Search for sea, haenyeo, or badang in the Archive tab.
- 4Open Graph and Sources to show provenance and relationships.
- 5Finish in Studio by verifying entries and generating a lesson pack.
Rubric Alignment
Creativity
Autonomous preservation agents plus a community-facing studio: not just finding words, but turning scattered sources into living lessons.
Impact
Endangered-language communities, teachers, and heritage learners get archives, review workflows, and lesson packs from the same pipeline.
Feasibility
The demo works without API keys through a full Jejueo fallback dataset, while live services can power real discovery and extraction.
Technology
Next.js 16, React 19, Socket.io, Elasticsearch hybrid retrieval, Jina embeddings, Featherless.ai agents, and map/graph visualizations.
UI/UX
A clean blue interface with map browsing, live agent observability, searchable sources, graph exploration, and explicit human verification.
Impact Model
at-risk languages indexed in demo mode
Jejueo entries represented in run artifact
audio clips in preservation metrics