aarushyadav

Exploring that too as a next product, but probably a while down the line!

Most probably not possible to do this remotely, since this kind of RF needs to be transmitted through the face, and if you transmit them from too far away very few of them will actually take a path that goes through the face and comes back to your receiver.

They get modulated since your mouth is a complex dielectric environment. Tissue, water, and air have different dielectric constants (different permittivity to RF) and the motion of the muscles in this medium leads to interactions with the waves that fill it, changing their phase and amplitude as they propagate.

It actually can! Since the RF can see inside your mouth the system doesn't inherently need lip movements and can be fine tuned on what we call "internal speech". Note that this isn't the automatic tongue movements that occur when "thinking" words to yourself, but it's rather like mouthing with your lips closed. We don't often lead with this since it's weird for many people to do, but with a bit of practice, we have been able to get it down pretty well.

Yes, we have! The same kind of RF doesn't work in that form, since it is attenuated strongly over that big of a distance. But the other interesting alternative we tried it mmWave radar, RF at about 60 GHz, the same band as 5G. This is much more similar to conventional radar, in that it would bounce off the face rather than passing through it like our current system. We found a few problems with this: 1. It faced the same issue as cameras. mmWave radar has a very low penetration in tissue, so it basically just bounced off the mouth. This only let us see visemes, which aren't enough to get to a low error rate as complete phonetic decoding can. 2. While mmWave radar can detect skin movement as small as a few microns via Doppler shift (given it's small wavelength), its spatial resolution with most systems is quite low. For example, it can tell that something moved 2 microns but within a 5cm x 5cm block. This coarse information makes it very hard to decode

That's why we weren't able to make something that can mount to your laptop. We're trying to make a necklace form factor at some point in the future though, but can't give a date on this!

Lip reading is one of the other modalities we tried as well beyond IMU, EMG and RF! We worked a lot on fine tuning Auto-AVSR, one of the most commonly used open source lipreading models. It's very enticing to work on, since it gets to about 17% word error rate, which looks good enough that you would think working on it further would get it down to product quality, but we found there's an information-theoretic limit on the modality. Speech has 40 phonemes, but only about 12 to 15 visemes (externally visually separable phonetic states). For example P and B look very similar from a camera. This causes the scaling laws on lipreading to flatten out at an error that's too high to be product ready but just low enough to make for a good demo. Our original plan was actually to ship a lipreading software to see if people would want this behavior, but we realized you couldn't make lipreading work!

Yup we worked with IMUs as well when we started this. They measure aggregated face motion, so cannot distinguish between fine phonetic classes, only broad words. The open source demo got to a 5 word classifier, I believe. We tried a lot to make IMUs work too, in various form factors, but the signal doesn't contain enough information for open ended decoding beyond classifying between a few words. We would have loved to be able to just push software!

I built a tool called Counsel that connects to your product and revenue stack (PostHog, Stripe, GA4, GitHub, Linear, Gmail, Slack, Notion, etc.) and runs a continuous analysis loop to surface insights, anomalies, missed opportunities, and suggested actions.

The goal is simple: founders often have the data but not the time to dig through it. Counsel tries to close that gap by running a weekly insight cycle. It generates hypotheses, tests them across your data sources, and produces a ranked list of findings. Examples: checkout friction, retention patterns, activation dropoffs, failed onboarding paths, pricing bottlenecks, unhandled error clusters, correlation between support volume and feature releases.

Right now we are running a private pilot with a few early stage B2C SaaS teams. They keep telling us the value comes from the insights they were not expecting. Small things they did not know to look for.

We are still early but would love feedback from HN.

If you want to try it, we are opening a small set of new pilot spots. Happy to share the technical details, architecture decisions, or lessons from building multi source agents if that is useful.

Link: https://www.withcounsel.co

Happy to answer any questions.