Show HN: LTE-connected IoT module with remote programming and NL data analysis

Show HN: LTE-connected IoT module with remote programming and NL data analysis(youtube.com)

27 points by siliconwitch 1 year ago | 25 comments

Hi HN! I've been working on this IoT platform that aims to simplify deploying remote sensor networks by combining pre-configured LTE hardware with a cloud platform for remote programming and AI-based analysis.

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!

flyingcircus3 1 year ago |

The remote programming seems like a novelty aimed at hobbyists. If I have a deployment of remote devices, large or small, iterating on code on a device in the field is a recipe for disaster. How will you track the different versions of code on the various devices you have changed individually? If you need to directly change the code on a field device, doesn't that imply that you are unsure that the changes will work?

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.

siliconwitch 1 year ago | |

The way the scripting engine is designed is such that it’s not really a problem to iterate in the field. If your code has an error, simply adjust and restart the script. Sure if you have the device hooked up to motors which could break things then of course you might want to do that in a controlled environment, but for applications like sensing that’s seldom a problem. The benefit you gain here is a very fast iteration cycle where you can easily test against real conditions, rather than in a lab where it might be hard to validate things. No need for heavy firmware testing, rolling out OTA updates, or worrying about bricking devices

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

usg7 1 year ago |

As this is going to be best use case for remote devices without human intervention, battery life or power source will play significant role in this. so as it's using LTE how power consumption is going to be (for places with less availability of electricity) any future plans for solar integration?

siliconwitch 1 year ago | |

We’re actually looking at a few energy harvesting ICs designed for solar. They usually charge up an onboard super capacitor that can be used for infrequent transmissions. If it works out, we might add it into a future version of the board, or as an add on board

klinquist 1 year ago | |

They mention it is battery ready, small solar panels w/ battery charge controllers are very plentiful in the market today. If this is anything like competitors (Particle), you can put the devices in a very low power sleep state and have them wake up on event or on a time schedule to connect to LTE and send data.

RickS 1 year ago |

I started out very interested. I tried the particle photon when it first came out years ago, and this feels like a spiritual successor, somewhat. Good demo video.

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.

xtiansimon 1 year ago |

Not a hobbyist in this area yet, but I’m curious. Hehe. The use cases I’m interested in are weather and water monitoring for fresh water fisheries. And in a completely different domain, I’m interested in noise analysis overnight. My neighbors are super noisy, and it’s difficult to demonstrate the breadth and depth of the disturbance to neutral parties.

brk 1 year ago |

Along the same lines, I've been working with a few of these modules and they are nicely engineered overall: https://quickspot.io/

h1t35h 1 year ago |

What do you mean when you say consistent response is a problem ? Is it that each time the response varies or that the response cannot be relied upon.

Have you tried fiddling around with temperature or SQL specific finetuned models?

siliconwitch 1 year ago | |

We haven’t tried SQL specific models actually. That’s quite a good idea. Will try it out.

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

999900000999 1 year ago |

I was trying to build an LTE solar powered phone of last resort.

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!

siliconwitch 1 year ago | |

It’s a massive rabbit hole! Thankfully LTE Cat-M1 that we use is much simpler than full blown LTE, but even still, the most iteration we’ve had to do is around the antenna system. Huge number of frequencies and bandwidths to support and much of the information is quite difficult to track down. Thankfully with the help of the SoC maker, antenna maker and SIM provider, we’ve just about managed to figure it all out

dogma1138 1 year ago | |

How is it better than having any phone + a backpack solar charger?

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.

999900000999 1 year ago | | |

It's more like if I get lost in my main phone battery dies or something or gets broken, I want a way to call for help.

audunw 1 year ago |

I don’t see any mention of what LTE SoC is used anywhere? Even if you’re abstracting away the hardware in this application, isn’t it relevant for many potential users what the chip is?

siliconwitch 1 year ago | |

It’s the nRF9151 from Nordic. We’ll include a detailed section on the SoC and a full schematic once the documentation is up

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

hakaneskici 1 year ago |

This looks very interesting, thanks for sharing. Have you started delivering any developer kits?

siliconwitch 1 year ago | |

Not yet, but next week we should be getting a batch of PCBs with better tuned antennas. Fingers crossed if the performance is good, we will send them out to test in different regions.

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!

klinquist 1 year ago |

An add-on GPS would be helpful for all sorts of remote telemetry applications.

siliconwitch 1 year ago | |

Yep! GPS is already built in