I am building Brandplane , brand-native voice agents for marketing agencies. Each client's brand becomes a LoRA adapter plus a cached context block, so one AMD MI300X serves dozens of brand-tuned agents at once. Speechmatics for real-time STT and turn detection; a multi-tenant vLLM server on ROCm as the agent brain. Who I need: Hacking solo, but open to one collaborator on the voice pipeline or demo/frontend side. I'm covering inference, LoRA training and serving, and benchmarking
.png&w=256&q=75)
Software Developer