first pass at personalizing Astro (#2)

* front page and site config updates

* updates the projects page

* cleanup

* updates site logo

* a bunch of small tweaks

* updates first note to announce release of invocate

* adds a note about the demo

* adds demo to the project page

* remove regular blog post section for now

* updates to the invocate post

* minor fix

* working on content

* adds some images

* minor updates to posts

Author: FredworkLemmas

public/big_top_tent__sm_trans.png

@@ -39,7 +39,7 @@ function formatCanonicalURL(url: string | URL) {
 <meta name="generator" content={Astro.generator} />
 
 <!-- Low Priority Global Metadata -->
-<link rel="icon" type="image/svg+xml" href="/favicon.svg" />
+<link rel="icon" type="image/svg+xml" href="/big_top_tent__sm_trans.png" />
 <link rel="sitemap" href="/sitemap-index.xml" />
 <link rel="alternate" type="application/rss+xml" href="/rss.xml" title="RSS" />
 

public/sunrise.jpg

@@ -48,7 +48,7 @@ function getYear(date: Date | string | number) {
     </template>
     <li v-for="(post, index) in list " :key="post.data.title" mb-8>
       <div v-if="!isSameYear(post.data.date, list[index - 1]?.data.date)" select-none relative h18 pointer-events-none>
-        <span text-7em color-transparent font-bold text-stroke-2 text-stroke-hex-aaa op14 absolute top--0.2em>
+        <span text-7em color-transparent font-bold text-stroke-2 text-stroke-hex-aaa op24 absolute top--0.2em>
           {{ getYear(post.data.date) }}
         </span>
       </div>

src/components/BaseHead.astro

@@ -1,8 +1,20 @@
 ---
-title: Note Title
-description: Your blog description, which is long text, can be an introduction to the post or a paragraph of the post.
-duration: 5min
-date: 2022-12-01
+title: Invocate
+description: I've released a wrapper around invoke that makes namespaces simpler.
+date: 2025-08-08
 ---
 
-Use [Vitesse Them for Astro](https://astro.build/themes/details/vitesse-theme-for-astro/) to start writing your blog posts.
+![](images/velocity.jpg "Velocity (the cat) wants to discuss his portions.")
+
+I just released [Invocate](https://pypi.org/project/invocate/) which is a
+packaged-up version of a wrapper I wrote a while ago to make namespacing with
+Invoke tasks a bit easier to work with.
+
+It's a huge improvement over what
+I've been doing, which is dragging around a collection of aging python code outside
+of a proper package.
+
+It also includes some (breaking) changes that I've been wanting to make for a
+while that will make it easier and more intuitive.
+
+And there are docs!

src/components/ListPosts.vue

@@ -0,0 +1,84 @@
+---
+title: "Demo: Pgvector"
+description: A vector database implemented using Pgvector and PostgreSQL.
+date: 2025-08-20
+---
+
+# Terra firma: playing at scale
+![](images/cat_creepin_outside_window2.jpg "Foxy (the cat) creepin' outside my office window")
+
+## My last demo was impressive, but pitiable in a lot of ways
+For the past year or so, I've been thinking about how People can benefit from
+LLMs and I've been noodling out a design for sharing context in an interesting
+way with friends, coworkers, and customers, but I've hardly touched any code
+that actually does anything interesting with an LLM or any other more typically
+AI-adjacent construct.
+
+But, just before that, I did a code up quick demo that showed off a simple RAG
+pipeline.  It worked remarkably well but it was dead simple: a lightweight
+model, a Chroma vector store, and some custom chunking code.
+
+A few weeks later, I built something similar at work to mine Basecamp
+conversations for support information.  Again, though just a demo, the
+results were pretty badass.
+
+## There were some pretty obvious scaling limitations:
+* at some point, I figured I'd want to put so much data in the database that it
+  wouldn't fit in memory and Chroma was an in-memory vector database.  I want to
+  see the day when we can search curated libraries that house vast amounts of
+  text, so persistent, non-resident, non-super-expensive storage is crucial.
+* the model I chose didn't fit in the VRAM so I had to run it with the CPU which
+  made it pretty slow (but not terrible really).
+* it had to download the model every time it ran.
+
+## Scoping out the future
+Because I know I'll be crossing these bridges at some point, I've been eyeing
+a bunch of answers to the demo's shortcomings.  PostgreSQL is an easy choice if
+it works since I've been using it for years.  Caching the model is an obvious
+upgrade too.
+
+Docling was a bit of an unknown, but it performed admirably as did vLLM in a
+Docker container, once I'd upgraded my drivers to the 580 version.
+
+## This demo
+[Demo: Pgvector](https://github.com/FredworkLemmas/demo_pgvector) is simply a
+proof-of-concept that shows off the same sort of RAG pipeline and query solution
+I'd built in the past, but with some enhancements:
+
+* A PostgreSQL-backed vector index, removing in-memory constraints.
+* Docling for cleaner chunking strategies.
+* vLLM with model caching, which makes small models easy to run repeatedly.
+* A Dockerized GPU environment, which turned out to be easier to configure than
+  in the past.
+* EPUB ingestion, Project Gutenberg unlocked!
+
+## Reflections
+* Embedding dimensions are strict: mismatches are non-negotiable...it's a choice
+  that's made when the DB table is created.
+* Model quality has improved: Qwen 1.5B was unexpectedly strong for its size.
+* Search thresholds were surprisingly low: semantic similarity scores were far
+  lower than I expected, making me wonder if it's actually possible to set that
+  as a constant.  it may need to reflect the content somehow.  and the oddly low
+  number also makes me think I should be baking in some sort of full-text search
+  (which happens to be pretty easy with postgresql).
+
+## Closing thoughts
+There were no "Eureka!" moments with this demo, but it was pretty easy to get to
+where all the moving parts were in place and working and the future is bright:
+
+* the AI assistant in PyCharm was super-helpful with some key bits of this
+  effort.
+* with a vector database that can scale beyond a machine's RAM capacity, a
+  surpisingly capable but smallish model, and some solid caching options with
+  vLLM, it looks like reliable performance on modest hardware is indeed
+  possible.
+
+## What's next?
+* There's a lot to be done in the context department - searching chunks is cool
+  but it's a subset of what a real-world use case will need.
+* It seems like there are some good ways to integrate MCP capabilities.
+* I'd like to see bigger models and multi-modal I/O.
+
+ 
+# Additional Notes
+THERE ARE NO TESTS!!  HERE BE DRAGONS!  RUN AWAY!