dhicks

Any update? I couldn't get our Smoothwall device to recognise more than one logged-in user at a time via RADIUS. I've now added idex (JSON packets) to the client, that does now seem to be working - I've rolled out my client to some test Windows and MacOS machines, we'll monitor over the next couple of days to see if there are any problems. Are Smoothwall planning to fix their own client to handle GCPW-style logins?

I think there was just a "publish" option within NotebookLM somewhere. It's not available on Google Workspace Education, only on personal accounts.

Hmm, interesting. I'll try and give that a go at some point and see what happens.

Depends on exactly what services you use and your current setup, but probably very do-able. We've removed our Windows Domain Controller and file servers and now use Google Drive / Workspace instead. We are fortunate in having good, reliable, fiber-to-the-premises connections - we actually have two, running via separate routes, and Google Workspace has proven very reliable (so far!). I think several schools were impacted by Azure outages a month or two ago, but that sort of thing is bound to happen at some point, just as long as your organization understands that sort of thing is out of your direct control, and overall uptime has been much better with cloud services. We still have on-site servers / controllers running our phone system (3CX), door access (Paxton), wireless (UniFi) and print (Papercut) - those things all relate to on-site hardware, a pure all-cloud solution is going to be hard. We run the servers for that lot as a bunch of virtual machines on a single Windows server, with another Windows server on standby and backup daily using Veeam. Still a reasonable amount of on-site hardware, although I suspect the actual servers needed now if we were buying new wouldn't be much beyond a standard Mini PC. We also have a separate Synology backup appliance which takes backups (every half hour?) of our Google Workspace, so we do still have an on-site backup of our files. We also run a remote desktop service for those users (teachers) who need specialist software (Timetabler) or just can't get the hang of not using Windows - I think we're now down to half-a-dozen users. That service could just as well sit in the cloud somewhere, it's just that we have the on-site hardware otherwise spare so we might as well use it. We have an on-site Smoothwall device for filtering, we are currently having issues getting it to recognize users logged on to Windows and MacOS devices - Azure AD joined devices might be better supported. Interestingly, because we have an on-site server cupboard, complete with two internet connections, two separate UPS-backed power supplies and cooling (and backup cooling!), we currently probably have a fair bit of spare capacity. That be useful to move to doing some AI processing locally.

There's a free version of NotebookLM available for users with a personal Google account - it sounds like your users are probably using their personal Google accounts to create NotebookLM notebooks. You probably don't want to allow that to happen for your school's data as you don't know where that data is going to go - it might get used as further training data for Google's AI models. Interestingly, you can make personal NotebookLM's public so that anyone can come along an interrogate your data, which could be a useful feature for something like a set of policies on a website, but not available for the Google Workspace version of NotebookLM.

You could look at Microsoft Power BI: https://www.microsoft.com/en-us/power-platform/products/power-bi Or the Google equivilant, Looker Studio: https://lookerstudio.google.com Both are free to start using, although Power BI costs if you want to start sharing dashboards you've created with other users. Another option is to write an application to process the data you want into a set of static web pages you can share how you like - I'd probably go for a Python script outputing HTML / Javascript using a pre-made graphing library. A year ago that might of been a bit much, but I suspect it's the kind of task that might now be a good fit for an agentic coding agent.

We have had a no-phones-in-school policy for pupils for several years, along with a no-phones-visible policy for staff. Younger pupils (Year 7 and below) don't tend to have phones, but for the seniors who walk home themselves they hand their phones in to the front desk at the start of the day and collect them at home time. When we had separate premisis for the senior school we had solid metal laptop charging trolleys with an individual shelf per laptop, they could swap their phones in (on charge) for the laptop at the start of the day and swap them out at the end of the day, that worked quite nicely. We're a private prep school, we have the staff numbers (teachers, TAs, front-desk staff) to be able to supervise all of this, it does need appropriate staff supervision and, of course, storage / charging space. Also, our children are all affluent enough that theft of phones hasn't really been an issue - each class currently just stores their phones in a metal drawer and rummages around to find theirs at the end of the day. The no-phones-visible policy for the staff is part of the reason we went for physical USB 2FA devices rather than phone apps. That has worked out okay, I think helped by the USB key also being the door access and printer fob, so it gets used on a daily basis.

Good point - we found that with the Ricoh synthetic media stuff it looked great (you can laser-print a full-colour A4 laptop cover label) but it did wipe off with some of the cleaning stuff we used.

You can buy it direct from Ricoh supplies, although annoyingly it seems you need to log in to their site to actually browse through their products. The thing you're actually searching for is probably "Ricoh Pro Synthetic media" - it's available in A4 and A3, and both opaque white and clear and self-adhesive or plain. You can put it straight through a standard laser printer with good results.

My guess is that you are unlikely to find a free / reputable tool similar to NotebookLM that includes free back-end processing, as AI computation can get expensive, and in the case of a school, you'll want to be sure you know where any data you put into the system is going and how it's being used. However, if you are able to host your own AI models (if you have on-site or otherwise servers somewhere) you could probably do the back-end bit on those. Hugging Face is the place to find self-hostable AI models: https://huggingface.co/

We found the self-advesive version of Ricoh's printable plastic A4/A3 "paper" actually stayed on quite well. For our first batch of Chromebooks we had each different room have a different image and had that on the front of the Chromebook and on the storage buckets they were meant to match up with. Worked okay with some of the junior classes, but by the time it got to about Year 6 they were just jumbled together. The stickers stayed on quite well, though, and indeed were quite tricky to remove if we wanted to switch devices around - which of course we did, which made the whole thing a bit more bother than we would have liked.

Outside my school work, I built and maintain a site where all the site's content is contained in a folder structure of Word and Excel documents. It means the end user can very easily edit content in nice, familier tools - I handle the structure of the site, they just need to be able to edit Word / Excel files. I'm using Pandoc to convert Word files to Markdown and Excel to CSV for further processing. The files are held on a Google Drive folder and edited using Docs / Sheets, but are actually maintained as Word and Excel files just for wider compatability. For Word files, for proper conversion into Markdown it's best to use the outline tools in Word (or Docs), where you use the various Headings to properly structure a document. Inside Word / Docs this makes it possible to do things like automatically create a table of contents, so it might be easier to explain to the end user like that. You might need an intermediate processing script that takes a single Excel / CSV file and produces multiple files from it for placing into WordPress. If you're writing a Python script, the very widely-used Pandas library can read Excel fies, that might be the best tool for handling your data. The Markdown files produced by Pandoc are being used to build a static site with a static site building tool (Jekyll / Hugo), I don't have much experience with WordPress but I would think there'd be a way to haul Markdown into WordPress. Possibly useful / related is that in Google Docs (and I'd guess there's an equivilant for Microsoft O365) you can "publish" the PDF version of a Doc as a PDF just by pointing people at the appropriate URL. This means you have just one Doc file, with suitible view / edit permissions so that only your staff can edit it, but it's "published" as a PDF document that everyone (if you want) can view. Any changes people make to the Doc update the PDF - just what you want for things like policy documents, where you can have the editable version behind the scenes and a PDF version published for the public to view. I don't think I've tried the same thing with a Sheets file, although I'm guessing something similar might be possible.

Any progress / updates on this? We've had to spend time writing our own user-reporting client to report users via RADIUS, now I'm having to figure out how to stop your Chrome extension loading on some machines so it doesn't then override the RADIUS-reported username.

We also use a simple on/off timer switch. We did use a Raspberry Pi a while back, the v3, when it could actually power itself from the USB port of a smart TV. Newer versions are a bit too power-hungry, though, and we've found managed ChromeOS devices (a £200 mini PC from Amazon with Chrome Flex installed, then managed as a ChromeOS kiosk device) to be more resiliant to having their power suddenly cut every day. We've not had any issues with our standard 65" Samsung TVs (they've been in our lobby for a couple of years), but honestly at education prices it's not much more expensive to get a touch screen, which are designed for all-day use and tend to be a bit more robust, so if a child does bash their bag / an elbow into it on the way past they'll probably just bounce off. We either use a simple Google Slides slideshow for our displays, or my own custom web page that can display content. For our lobby display screens, if they were touch-screen I think a "click here for more info..." type kiosk might be nice.

To follow on from the comments about laptop charging ports, we use Chromebooks, most of which have two available USB-C charging ports, one each side. Very handy for if one does break for some reason, and does mean laptops can be stacked either way around, which makes putting them away a bit easier. I'm guessing similar hardware should be available for Windows devices. We use Lock-n-Charge cabinets, which have proven pretty robust. These are the ones whith the 5-device carry buckets inside. We find, with primary-age children, 5 devices is actually quite heavy and difficult to carry, so we've tended to swap them out for dividers. They provide standard dividers, but we've found metal lever-arch file stands work well, allowing us to fit more devices in per cabinet. We've swapped most of our individual USB-C chargers out for multi-way USB chargers, to which we plug in standard USB-A-to-C cables that are cheaper to replace than a whole charger. They also draw less power / ground leakage(?) than multiple laptop power bricks, making it less likely to trip the ring main out.

If you are okay with a bit of hosting / editing things yourself, I have a basic "slideshow": https://gemini.google.com/app/37b8c937bdd9dac1 That's a server-side Python script to prepare various resources, and an HTML (and JavaScript) web page that displays the resources. Server-side you could use rclone or similar to sync files accross from Google Drive. The web page should, hopefully, be usable stand-alone if you want to do your own server-side script.

From a quick test, it seemed accurate, although this is the kind of thing that really needs testing in practice - it answered my bunch of test questions okay, but how it handles the sort of thing users throw at it I'm not sure. The first problem we encountered was collating all the appropriate files to put into the Notebook - some are in one folder, some in another, this is the kind of thing we'll need to take a bit of time to have a proper sort out with. I was quite impressed by the "Further questions" feature - at the bottom of a generated answer, you generally get two or three "Have you also thought of asking...?" questions, which was actually quite useful for the policies Notebook - you can browse though linked questions better than trying to read a 100-page policy document.

We currently have inspectors in, doing a pre-inspection run-through. Everyone is being reminded to look at our policy documents, which of course are impractically large to simply read through. A NotebookLM of all the policy documents, shared around staff and available for them to ask questions of as needed, might be a really nice use-case. The same, possibly, with our technical support documentation. We have a bunch of (hopefully well-written) documents describing how to go about various tasks / processes. Piling all those into a NotebookLM and, again, letting the staff ask questions potentially gives us a technical support assistant. A problem with NotebookLM seems to be that it won't scan for changes to a folder of documents - you have to manually remove the old document and add the new version. This looks like it should be scriptable, though, so we can maybe have something scan for changes daily and update Notebooks as needed.

I suspect this might be more brand / keyword recognition than any real preference for ChatGPT. Similar to what happened with Zoom during the pandemic - Google and Microsoft had more manageable solutions probably more appropriate for a school / corporate setting, but for some reason staff tended to talk about / try and use Zoom as it was the one they remembered. At some point we realised people were referring to Google Meet as "Zoom", and that, in their heads, "Zoom" was a generic term (like "Hoover") which just meant "video conferencing". We have a Google Workspace subscription that includes the more advanced features of Gemini, so we've gradually been trying out some features. Other AI tools are blocked - Smoothwall's "AI tools" category seems to be quite comprehensive. Google's NotebookLM (which allows staff to processes a bunch of documents and then interrogate those via a chat bot) looks useful, and I'm currently finding Gemini is better able to provide answers to technical questions about various products used in an education context than the average support person. I'm aiming to give an agentic tool a go, where you can give an AI agent access to files and a workspace of some sort to help complete some tasks, but I'm keen to get a controlled, limited environment set up first before having staff or pupils try and use that.

If you have a map showing you the positions and models of the lights, and you know they've all been set up already, you can probably just cycle through all the DMX channels and see which lights come on when you fade certain channels up. I've done that before at a school which had a bunch of lights on a three-story-high sports hall ceiling - rather than try and climb up to see the lights, we just started at DMX channel 1 and worked our way along, seeing which lighst came on as we went.

We're aiming to have guests connect to UniFi with a voucher (dished out to them by Inventry's integration with Unifi), then have UniFi inform the Smoothwall server. I've had a rummage around their documentation today: We don't seem to be able to set a UniFi Hotpot send a RADIUS packet unless we want to use RADIUS for authentication - we want to use a voucher. There doesn't seem to be a UniFi alarm / event defined for "Guest user has logged in with voucher code xyz". Therefore, the solution seems to be to have something poll the UniFi API every (say) 30 seconds, gtetting details of any vouchers used and updating the Smoothwall server via RADIUS accounting as appropriate. This means still having a local process of some sort running, but that can be a small application on a server, you don't have to have an Active Directory / NPS server on site. The polling client could (probably, I haven't checked) poll for new RADIUS users authenticated with the UniFi system and inform the Smoothwall server via RADIUS accounting.

Did you have a go at testing / implementing anything yet?

Just thought: could you have your BYOD users authenticate with your Smoothwall server rather than your UniFi system? Smoothwall handles cloud logins quite nicely (at least, our login-with-Google logins seem to be working okay on our Smoothwall appliance), I'm pretty sure Azure AD logins are equally easy to set up on Smoothwall.

Ah - a quick Google suggests: Possibly, UniFi might only pass Framed-IP-Address if it is handling DHCP itself - not sure if that's accurate, it seems a bit of an odd limitation.

Ah, this is something to test then, that might well be sorted by now - I'll hopefully have time to look at this on our system over the next couple of days. How expensive? It should be possible to set up a RADIUS server on a cheap cloud-based VM (£5-a-month sort of cost), but that would, of course, need setting up and maintaining.