Gridspace develops technology at the intersection of voice networking and speech machine learning for real-world conversation. We are starting with phone work.
The company is run by its founding team. They built systems for analyzing massively concurrent voice streams in real-time with ML.
Since this milestone, Gridspacers have pioneered new LLM techniques for live conversation search, real-time analysis and task-based spoken dialog systems. The team has also grown – thanks to close collaborations with the speech and language programs at Stanford and MIT.
Recognized by Gartner, HBR, NeurIPS, and CBSNews, Gridspace continues to advance the frontiers of agentic voice. For careers, find us at the Stanford and MIT fairs. Also watch our CS406 lecture from last Spring.
Since this milestone, Gridspacers have pioneered new LLM techniques for live conversation search, real-time analysis and task-based spoken dialog systems. The team has also grown – thanks to close collaborations with the speech and language programs at Stanford and MIT.
Recognized by Gartner, HBR, NeurIPS, and CBSNews, Gridspace continues to advance the frontiers of agentic voice. For careers, find us at the Stanford and MIT fairs. Also watch our CS406 lecture from last Spring.
Some technology highlights
Gridspace's entire voice stack SIP gateways, switching, transcoding, and neural analysis, is explicitly engineered for reliability, control and latency.
Gridspace Jswitch: Gridspace's voice inference infrastructure, which leverages specialized LLMs, reduces latency by loading and serving model directly on Jswitch, a neural voice networking switch.
Gridspace Findo: A highly scalable enterprise search and analytics backend, specifically for agentic voice. It is capable of handling trillions of tokens.
Gridspace Mixer: A domain-specific crowdsourcing platform for labeling challenging conversational speech data sets related to task-oriented voice agents.
Gridspace Petal: A massive parallel data processing engine for speech and language data. It reduces the length of processing jobs from weeks to less than an hour.
Voice shouldn't be the most expensive channel for businesses to engage consumers and the most frustrating for consumers to interact with businesses. At Gridspace, we believe better voice agents can help.
Gridspace's entire voice stack SIP gateways, switching, transcoding, and neural analysis, is explicitly engineered for reliability, control and latency.
Gridspace Jswitch: Gridspace's voice inference infrastructure, which leverages specialized LLMs, reduces latency by loading and serving model directly on Jswitch, a neural voice networking switch.
Gridspace Findo: A highly scalable enterprise search and analytics backend, specifically for agentic voice. It is capable of handling trillions of tokens.
Gridspace Mixer: A domain-specific crowdsourcing platform for labeling challenging conversational speech data sets related to task-oriented voice agents.
Gridspace Petal: A massive parallel data processing engine for speech and language data. It reduces the length of processing jobs from weeks to less than an hour.
Voice shouldn't be the most expensive channel for businesses to engage consumers and the most frustrating for consumers to interact with businesses. At Gridspace, we believe better voice agents can help.