Papers, awards, and talks — newest first.
Cost, latency, and where the data lives — the case for small, on-device models from our WigtnOCR work.
A 2B-parameter document parser distilled from a 30B teacher — ranked #1 on the KoGovDoc Korean government-document benchmark while running on a single consumer GPU.
WIGTN Flake turns Snowflake Cortex into a purpose-driven neighborhood-intelligence platform — five AI experts debate across four datasets.
Two parallel AI interpreter sessions deliver bidirectional voice translation over standard phone lines — 557ms average latency, zero echo-loop incidents. Accepted to ACL 2026 System Demonstrations.
Watch Cortex-powered experts cross-query four datasets and converge on a ranked Top 3 of Seoul districts.
Visual code refactoring with an always-on AI agent — point it at your dev server, rearrange components in the browser, and the source updates itself. Published on npm.
12 agents, 3 skills, and 17 design styles working together with team-based parallel execution for a 3–5× speedup.
WIGENT, a multi-agent debate arena, won the Grand Prize — built by 3 engineers in 3.5 hours.
A walkthrough of the dual-session architecture and software-only echo cancellation behind WIGVO.
An AI museum curator that narrates exhibits in real time via Gemini Live, with historical context and restoration visualizations.
