← All companies
Hopper
ActiveFast inference for voice
F26·Fall 2026·B2B·San Francisco, CA, USA·Team of 2·Founded 2026
About
The best voice model is small, specialized, and yours. Hopper post-trains STT, TTS, and speech LLMs on your production calls, serves them at 100ms to first token (GPT-4.1 is ~700ms), and keeps improving them against your live traffic.
Founders
Pavan Katta· Founder/CEOPreviously led voice inference and post-training at Salient ($60m series a from a16z, voice). EE @ IIT Madras.
Jashwanth Pedapudi· Founder/CTOFounder at Hopper. Previously kernels engineer at Meta Reality Labs. Electrical Engineering from IIT Madras.