Daz

Vox Commander is a real-time strategy game you play with your voice. No keyboard, no mouse — just spoken commands. RTS games are the most input-dense genre on PC: pro players hit 300+ actions per minute, every second a flood of clicks, hotkeys, control groups, and menus. If voice can replace mouse + keyboard here, voice can replace them anywhere. The system is three layers. ElevenLabs Conversational AI owns the voice — streaming STT, intent decision, and the XO persona that confirms every order in sub-second turn-taking. A Python tool host exposes ~25 typed intents (focus_fire, set_rally, hold_position, station_army, airstrike, toggle_grid, …) the agent can call. A custom C# trait inside OpenHV translates those intents into the same native game orders a mouse click would issue.

Submitted 14 May 2026

Creating anime music videos (AMVs) traditionally requires hours of manual editing — watching footage, finding the right scene for each lyric, and cutting on beat. This project automates the process: ElevenLabs generates the song with word-level timestamps from a text prompt, while turbopuffer's hybrid search (vector ANN + BM25) instantly finds the best matching anime clip for every beat from pre-indexed scenes. The result is a fully beat-synced AMV generated. Demo video: https://youtu.be/k597okobXsc Slides: https://tantk.github.io/amv-search/

Submitted 16 Apr 2026