The original Disinfodrome cluster was turned down at the end of January. The Dell systems were the last of their servers made with DDR3, so they were all 10+ years old, and nobody wanted to foot the bill for their hosting.
Now, thanks to startup duties, there’s a Proxmox cluster in my hut for the first time in five years, and I have some limited ability to deliver indexed data once again.
Come below the fold if you’d like to know more …
Attention Conservation Notice:
Pre-elderly document hoarder mumbles at length about semantic search and the implications of the advent of vector databases. You have been warned …
Infrastructure:
As I’ve mentioned before, thanks to Angels In America I’m getting a bit of a hardware refresh. The RTX 5060Ti sits here, awaiting a new workstation that will arrive this week, but cleaning while waiting, I found I had enough equipment to build a proper three host Proxmox cluster.
This will have one server grade workstation with plenty of storage and two small desktops so I can experiment with Proxmox cluster management, as well as getting familiar with Kubernetes horizontal scaling. The two small machines have enough disk that they can support Open Semantic Search deploys, and one of those two has already been claimed.
So there you have it. There’s on little machine sitting here and it needs some productive work. Who has a pile of documents they need to explore?
Operating Costs:
This stuff is cheap but it ain’t free. Here are some examples of things holding me up because they’re sitting in my cart rather than on my desk.
Implications:
If you’ve got an OSS server, that comes with an Apache Solr API. And just look at what MindsDB offers in the way of connectors.
The default vector database for MindsDB is Chroma, whose MCP server I forked to make Parabeagle. So it looks promising, but …
This is NOT a vibe coding job. OSS stopped with Solr 7.something and brand new MindsDB only expects to see Solr 9. There will be quite a bit of tinkering to fix this temporal disconnect.
Shall I fork OSS, which is now abandonware, and update this deterministic, semantic focused search tool to include artificial intelligence functions? Keep in mind that six years ago I waded through the process of hand building every single component of the OSS system. It’s all Python, as is Parabeagle, and it’s all vaguely familiar to me.
Conclusion:
The hours are long, the days are short, and I have miles to go each and every time I open my eyes. Even so, there’s a LOT of synergy here - old OSS, new Parabeagle, curation duties that are part of the startup.
I’m at the “realizing something is going to happen” phase of the process. I just have no idea what.
Please subscribe and or send me Amazon gift cards, so I can make it all happen :-)
