CHOOSEConsumer & creative
Jev-Powered Intent Classification for Voice-to-LaTeX
Vivek Karmarkar@vivekkmkpinn
1:22This project uses Jev for intent classification to map natural language input, processed by a denoiser LLM, to computational primitives for a simulated voice-to-LaTeX system.
I have been messing around with Jev @typesafeai and since it is good at decisions/classifications - I had an idea...
If you have some "intent" and that maps to a "computational primitive", then given the intent, choose the primitive is a classification task and given natural language input that's decomposable into intent atoms, you basically can have a "fast LLM + Jev" LOOP.
I applied this idea to a "simulated voice2LaTeX digital blackboard" that essentially simulates
live garbled transcription ->
runs denoiser LLM ->
split into atoms ->
map atom to primitive with Jev
Had Claude Code @ClaudeDevs @claudeai one shot this in about 4 hours with /goal /model claude-opus-5[1m] /effort ultracode problem_statement_latex_idea.md with a demo video made with @ElevenLabs @ElevenLabsDevs MCP
Check out the result @belltyler @Drewch
