Trying to gauge interest for an on device, low compute, deterministic NLU engine for voice control? For the past 2.5 years have been full time into NLP R&D with a focus on natural language understanding.
Not advertising, but first offering is live with details and demo at: Sophia: Home Assistant Edition - Sophia NLU
It’s a purpose built Home Assistant edition, and I’m wondering if there’s any appetite for a Balena edition? There’s a demo there along with link to example sentences to give you an idea of its capabilities.
Rust based, on device, low compute NLU engine that offers the fluidity of a LLM without the compute and requires only 180MB RAM, which can be decreased for a Balena edition. Handles custom vocabulary, multiple intents per message, ambiguity, contextual awareness, ~15ms latency, doesn’t connect to the internet and never calls home.
Fully deterministic, so 100% reliability with zero hallucinations or probabilistic mismatches. Previous generations of deterministic NLU are generally pre-defined sentence templates with slots, which is quite rigid and not very nice. This is the next natural evolution of that, allows far greater fluidity and noise while being able to infer intent based on context with great accuracy.
I’m curious, is there a demand for such a thing within the Balena eco-system? Is it worth the time and resources to develop out?
If this is something you could utilize sooner rather than later, feel free to e-mail me directly at matt@aquila-labs.ca and would be happy to get the ball rolling for you.