The core of our offering is a discrete sensor that leverages multiple inputs (primarily an imaging sensor + PIR-based motion sensing), which feed into a neural network model that executes inference directly on the device. This allows us to do powerful processing on inexpensive hardware.
Our machine-learning stack is built around Tensorflow, which we use in two ways: 1) for inference (we embed Tensorflow directly on a Raspberry Pi), and 2) training new models in the cloud. New models can be pushed remotely to the devices over-the-air to make the sensors “smarter”.
While our sensors are currently trained to count people, our vision is to evolve into a 100% passive "super-sensor" that can be configured to detect thousands of different types of events. Examples that we've explored include things like detecting falls (e.g. during an emergency), counting assets (equipment, furniture, cars), and monitoring equipment usage (for preventative maintenance).
We're happy to chat and would love to hear your thoughts. Some things we've worked on that might be interesting to discuss: rapid-prototyping for hardware (Raspberry Pis +ESP8266), machine-learning, computer-vision, building automation, BLE, B2B sales, keeping sane while drawing bounding boxes, or anything else that comes to mind!
We look forward to your feedback!
Dan + Kelby
Uptake is strong, as you say, because facilities management can benefit a lot from condition-based monitoring enhanced with ML.
Good luck - reach out if you want to chat,
Will
https://news.ycombinator.com/item?id=14947275
Tried to change the title of that post to:
VergeSense’s (YC S17) AI sensing hardware wants to reduce the usage of office space
but HN didn't let me add (YC S17)
- What about privacy, is filming workplaces in high resolution ok with customers, their employees, the law and unions?
- Your FAQ states that you are selling the whole package for a yearly fee. Isn't that quite a risk when the customer is using mobile data as connectivity and having devices in the field that can and will fail and have to be replaced? Do you pay then a contractor to replace a single hardware node at your customers location?
- Have you looked at warehouses as customers? I suppose real estate is their #1 cost center :)
- We include a gateway device with our product, and if anything goes wrong (sensor or gateway goes offline), we cover this as part of our service contract.
- Warehouses are another potential vertical, provided we have access to training data to train up our models. For example, if someone wanted to “count” things like boxes / forklifts / etc, our sensors can be configured to detect them.
Modulating heating / cooling based on the exact count can help cut energy consumption, sometimes by as much as 30% for commercial buildings.
ARPA-E (Advanced Research Projects Agency) recently put out a proposal for such a system - you can read more here if you're interested:
https://arpa-e-foa.energy.gov/
Do you mean each device gets smarter individually because the specific device learned more about the specific space? Or that there is some kind of supervised learning component where you would adjust the algorithm/model over time for every device.
Also, what this the average area the sensors cover
As for sensor coverage, we cover about 1k sqft per sensor (it'll vary a bit depending on mounting height - higher mounting equates to a wider area of coverage)