Since I’ve been overdue for another blog post, I wanted to quickly cram a few topics into a single post. Below are a few updates from the second half of 2024.1

Raivo OTP Cluster F***

Over a year ago, the opensource provider of One Time Passwords Raivo OTP was quietly sold to a for-profit company Mobime. Nearly a year passed before any updates were pushed by the new owners. Then shit hit the fan.

Panickedly, I scoured Reddit for recovery methods. Finding little hope and lots of rage (see pic), I came to terms with the reality of starting anew with a different OTP provider. I was relieved I wasn’t the only one impacted… A quick check of the GitHub repository’s reported 100+ issues indicated the world was indeed on fire:

Raivo OTP Github Issues Screenshot

One of the lessons learned for Open Source software is the implied warranty can drop like a hammer. I suppose, however this wasn’t a reflection of any inherent flaw in open source development, except that the original developer built such a good product that another developer was willing to buy out the project and take it a different direction like a shitty, ransomeware one. I should have migrated as soon as the project changed hands…

By now I’ve migrated to a different provider; 2FAS. Coincidentally, the branding and colorscheme is eerily similar to Raivo, but so far, I’ve been pleasantly surprised. And I’ve enabled backups. 😰

LLMs in Regular Life

In the fall I completed a course through Stanford Continuing Studies titled From Data to Deployment: Creating Production-Ready Large Language Model Applications and it struck me as paradoxically complex and simple at the same time. I’ll explain.

My main goal of enrolling–as a non-product-owner or software developer–was to more aptly identify strong use cases for LLMs and know when to avoid black holes spawned by the GenAI craze. 😵‍💫2

Learning how LLMs fit into the evolution of natural language processing and search techniques was hugely insightful, yet also humbling. The humbling part is that the power of these tools is limited to the quality of your prompt/query and the context or data to provide a meaningful answer. And they aren’t immune from hallucinations.

Now, as I chat about LLMs with friends and family more often, I realize plenty of simple applications exist for every-day life, merely by bringing together helpful context:

  • Help with brevity and clarity in written communication, i.e. emails
  • Summarize (or seek specifics of) long or unstructured documents, i.e. PDFs
  • Help with code: Google Apps Scripts for my Padres Ticket Manager, Home Assistant template sensors, or code troubleshooting.3

The real competitive advantage for LLMs in tailored applications comes down to heightening the awareness and domain of the model; tax or personal finance applications, for example, require tremendous detail to get personalized recommendations. This is why I’m so excited Home Assistant is bridging the gap between voice assistants and LLMs.

I am inspired by folks leveraging ChatGPT to get creative reminders/nudges in their mobile notifications through Home Assistant, including wry reminders or scolding you for increases in water or electrical use.

Blinds Go Smart

In a previous post detailing all my smart home devices, I made a forward looking statement regarding smart blinds. Well, ladies and gentlemen, the time has come. The verdict is in; they are awesome!

I saw a deal over the summer on Switchbot blinds. Again, I must stress the versatility afforded to me by Home Assistant; I’d never heard of or used any Switchbot devices prior. They also sell a home hub, but I opted to skip that and hopefully tinker my way to a generic controller for the given protocol (Bluetooth Low Energy or BLE).

After programming and calibrating my blinds on their rod using the Switchbot mobile app, I realized I didn’t have the proper bluetooth hardware on my mini PC for centralized control with Home Assistant. So I ordered 3x ESPHome low-energy boards and flashed the firmware through Home Assistant to use them to communciate with my Switchbot blinds.

Now, I have the blinds programmed to open on workdays at 6:45am (so long as I’m home), which is the same time my alarm clock radio goes off. So even if I hit snooze, I still am met with bright morning light. For non-workdays, they open a little later, and can be easily overriden.

Bonus points: I picked up a bunch of RFID tags from Switchbot, which I haven’t programmed but would like to incorporate into various chores around the house to help remind me if they’re over due; watering plants, replacing HVAC filters, and tasks with varying cadences. Another great protocol that is natively supported by Home Assistant.

Proxmox & TTeck’s Legacy

Late in 2024, one of the biggest champions of Proxmox VE passed away. His wife posted to his Ko-Fi page with the news:

Good afternoon! I am tteckster’s wife. I don’t have a clue if anyone will even see this because I’m not the computer savvy person that my husband was, but I wanted to try. I know that he posted an update regarding his health the other week, and I wanted to let you all know that he passed away a few days ago. If anyone sees this, maybe you could make a better post. Thank you for all that supported him. Angie

To consider Angie had little to no idea what TTeck was doing all those years, seeing the outpouring of love and appreciation for his contributions from hundreds of community members is quite special. I hope his family finds comfort knowing how impactful he was to so many.

I have been meaning to write a blog post reflecting on my journey with Proxmox VE. Nearly every one of the tools or projects I’ve adopted and come to rely on were made so accesible thanks to TTeck. RIP.

2025 Smart Home Resolutions

Finally, as I kick off the new year, I’m brainstorming different tools and projects I want to learn or incorporate into my smart home automation. This is more of a list of considerations rather than resolutions:

  • Home/Away/Guest Modes: I want to double down on automations for quality of life, but I need a failsafe way to disarm myself from my own arsenal of automations.
  • Nextcloud: I’m fed up with all the You’re almost out of cloud storage! spam and I am determined to DIY my own backups for infrequently accessed items like photos and videos, and, hopefully soon, network video recordings (NVR).
  • Paperless Post: I’m still trying to determine the right way to digitize all my product manuals for retrieval augmented generation (RAG) with LLMs. Which brings me to
  • LLMs IRL: looking for the right medium to leverage the power of LLMs, in notifications like announcements, reminders, as well as multi-modal without stranding me on an island.4
  • Bonus: Thread/Matter or Presence Detection: these are some of the latest innovations in smart home sensors and device networks. I’m thinking of testing the waters…

Cheers to a year full of projects, blog posts, and adventures!

  1. This style of post draws on inspiration from one of my favorite columnists, Matt Levine of Bloomberg. His writing style is S-tier in his domain. Highly recommend. ↩︎

  2. In my professional life, I frequently see colleagues or customers refer to GenAI, when they mean other types of machine learning. Seeing this gives a strong When the only tool you have is a hammer, every problem begins to look like a nail vibe. Or combining other ML-terms into redundant AI phrases. Like naan bread, or mein noodles. ↩︎

  3. Talk about meta, I’ve been leveraging ChatGPT to help me roll out some small feature enhancements to my blog. Rather than pouring over HUGO documentation, I simply provide context of what I’m after and ChatGPT helps with shortcodes, partials and more to make any features highly scalable. Best of all, I can ask for more explanation of any parts to grow my understanding of HUGO fundamentals. ↩︎

  4. An Artificial Intelligence silo is just as bad as any of the other silos I’ve tried to combat in my smart home environment. I don’t want LLMs confined to one narrow aspect of my life. I want the flexibility to bring highly-personalized data from various sources so it can peform with the right level of context: my calendars, weather, traffic, sports, news and more. ↩︎