The main challenges I'm trying to solve are:
1. Eliminating infrastructure setup headaches for IoT deployments 2. Making remote programming and debugging practical for devices that might be difficult to access physically 3. Using natural language for analyzing sensor data, and possibly taking actions based on the analysis
I'd really appreciate feedback on:
1. Is this approach to IoT development interesting to you? 2. What use cases would you want to explore with this kind of platform? 3. What concerns would you have about adopting something like this? 4. Could anyone recommend workflows or tools for making the AI agent more reliable? Currently using LLMs to generate isolated SQL queries to extract data, but ensuring consistent responses has been challenging.
Thanks for any thoughts, and feel free to ask any questions about how the hardware or platform works. Happy to dive into the details!
If the code needs to change, that is much safer to do in a controlled, low risk, nonproduction environment, where you will very likely have development hardware that has been optimized for ease of development/debugging. Once changes are tested and otherwise validated, an over the air update process would be used to send that new firmware image to the device. This way, your devices all can be on a single version of the firmware, meaning you can scale the deployment of that update to 10, then 100, then 1000 devices, etc.
In general I agree with you for complex projects, but these are typically where you can afford to have 2-3 engineers spend a year developing a specialised solution. Building infrastructure to safely roll out updates, managing data plans, handling compromised devices. All of this becomes a massive amount of work before you’re actually solving a problem for your business
The target audience isn’t necessarily engineering companies, but rather those that might have a few tech generalists. Their goal simply being to add some kind of sensing that might transform their product or service
Various thoughts:
1. Consider defining best practices for solar and selling harder around that. I've got enough use for the board in hardwired environments, and can otherwise solve outdoor power issues, but having a recommended solution that I can trust to be solid would go a long way. "Using panel of X strength and battery of X capacity, get performance Y in Z conditions", etc.
2. I would probably preorder some of these, except the pricing is omitted. I have no idea how much these would cost. If they're 5 bucks, I buy a bucket. If they're 30, I buy a handful. If they're 100, I probably skip it. Etc. Withholding the price is a red flag, and I wouldn't share my email with an org that saw fit to withhold this info as the opening act. (edit: looks like the S1 is $55, so presumably this is more)
3. The LLM/agent aspect has no appeal. Assuming those costs are passed on to me / baked into every unit, the inclusion of AI is strictly downside.
4. It's not clear whether there are ongoing SaaS/storage/subscription costs associated with this, or what they would be. The FAQ suggests that forwarding data outside of the cloud will be restricted to enterprise-scale customers. This is also a red flag.
I think this can be successful in spite of all that. Particle definitely leaned away from hobbyists and into the larger ag-IOT market early on, presumably following money and stability. Totally understandable.
I guess my overall feedback is this: be upfront about the pricing and restrictions in a way that lets guys like me filter ourselves out upfront, instead of getting our hopes up. I've got a maker-tier budget, value data freedom, and am subscription-averse. It took me quite a bit of time and digging to uncover all the details in this comment, and I wish I hadn't spent the time. There are a ton of customers who are going to be totally cool with a $60+ board and a *0,000/mo contract for data forwarding. Court them directly.
Of course, I'd love to be wrong. If this is a $30 board with nearly-at-cost cloud storage and no REST data forwarding but yes bulk JSON download, then whoa, fuck yeah. Shout that from the rooftops too.
Either way, disambiguation would help.
We’re modelling the hardware pricing such that the cost of the AI isn’t baked into the board cost. I understand that for some users, they might not necessarily need these features and it makes no sense for them to have to pay for it.
I’m expecting that we can have the boards competitively priced around the ~$70 mark, but often the distributors will try to price it in line with similar products they carry. I didn’t want to mention it in the video as we’re still negotiating this with them.
Going up from there, the subscription pricing will be split into several tiers that allow for higher data usage, AI usage and the ability to connect higher numbers of boards to one deployment.
It’s interesting that you mentioned that you’d like the data forwarding as a base feature. Our initial thoughts around that was to fulfil the requirement for companies who need to own their data. Ie we can’t store it for them. To be clear, you’ll still have access to the data via an API even in the free tier, and you’ll also be able to send data to devices via a REST API. It was just the live data forwarding from devices to your server that was intended to be limited to companies
Have you tried fiddling around with temperature or SQL specific finetuned models?
Yeah it mostly seems to be inconsistent at generating the SQL queries themselves. I guess my question was if this is even a good way to do it in the first place, or if we’re better off trying to analyse the data in a different way
After looking into it, LTE makes this nearly impossible to do economically.( Plus there's like 4 different types of LTE depending on county.)
I really want energy coms devices!
If you are looking for a doomsday scenario then LTE isn’t the way to go, handheld radios is the way to go.
In virtually any large scale disaster scenario cell networks are one of the first things to go, they get overwhelmed and if there is a power loss then cell towers go down.
The platform as you said aims to abstract away a lot of these details (though they’ll still be available if needed). Each of the IO ports will be configurable as GPIO/Analog/SPI/I2C, etc. Since these are all accessed via the Lua API anyway, we expect folks to not have to dig through the SoC documentation in order to figure things out. All the relevant settings for each mode will be available through the API. But yes, if you need more info, then it’ll be available in the docs
If you have an nRF9151-DK though, we could provide a firmware image and SIM for you to try it out. Once you flash your board, you can use it with the platform just like I did in the video
Let me know if you’re up for trying it!
