No matter how shitty today was, at least I wasn’t patching php
Michael Cizmar
@michaelcizmar.bsky.social
technology veteran and entrepreneur who has founded several companies, specializing in enterprise search, AI solutions, and transformation.
This week's PLaiGROUND weekly: - #Jev in the wild - @NVIDIA buy back - @Google Meet meets Teams - Nrtsearch from @Yelp - turbopuffer agrees to what we've said all along, it's not a vector database - and more.... plaigroundaiadvisory.substack.com/p/the-weekly...
The Weekly Plaiground - 2026.10.03
JEV is this quarter’s Copali
plaigroundaiadvisory.substack.com
#OpenCrawling released the @ApacheSolr Output Connector (oc-solr-output-connector)! OpenCrawling now natively streams enterprise document repositories into #Apache #Solr 10.x and 9.x opencrawling.org/blog/introdu...
Solr 10 Dense Vector Search & HNSW Quantization: Announcing the Apache Solr Output Connector in OpenCrawling
Discover OpenCrawling's native Apache Solr Output Connector (oc-solr-output-connector). Built with SolrJ 10 native JDK HTTP Client, solr.DenseVectorField, HNSW quantization, efSearch tuning, and REST ...
opencrawling.org
The Frontier Has Moved A Lot Closer...Almost Local. I ran to code reviews on a workspace and guess what...both missed things. But with the local model it barely cost me anything and the frontier model...a lot. #Qwen 3.8 #Claude michaelcizmar.com/blog/2026/08...
The Frontier Has Moved A Lot Closer - Local Coding Agents Catch Frontier
It's very difficult to beat the performance improvement that comes from properly setting up your analyzers in an industry standard manner.
michaelcizmar.com
This week's Plaiground Newsletter: + Benchmarking + Benchmarking + Benchmarking Also: + Hypothetical Document Embeddings + #Jingra from @elastic + @Google's Fab Tabfm model + @couchbase gets off the couch plaigroundaiadvisory.substack.com/p/the-weekly...
The Weekly Plaiground - 2026.07.07
All solutions are equally poor if you do not benchmark
plaigroundaiadvisory.substack.com
From F-Droid to independent devs, the entire Android ecosystem is under threat from Google's new mandatory registration decree. See why 67+ orgs are saying no: keepandroidopen.org
Keep Android Open
Your phone is about to stop being yours. In September 2026, Google will block every Android app whose developer hasn't registered with them.
keepandroidopen.org
The AI Bill has arrived. Plus #HermesDesktop #PaddleOCR VL-16. The Weekly Plaiground - 2026.06.10 open.substack.com/pub/plaigrou...
The Weekly Plaiground - 2026.06.10
The AI Bill Has Arrived
open.substack.com
Need a spreadsheet to calculate your #Elastic search cluster sizing for #hybridsearch #RRF I've put together that I have used a couple of times with a little better formating: open.substack.com/pub/plaigrou...
Sizing Elasticsearch for Search
The storage and compute assumptions most teams get wrong — and a calculator to help you get it right.
open.substack.com
Haystack US 2026 Session Videos have been released. great sessions to learn about #AI / Search / #Agents: www.youtube.com/playlist?lis...
Haystack US 2026 - YouTube
Haystack US 2026 took place in Charlottesville, Virginia on the 6th and 7th of May 2026. Haystack is the conference for improving search relevance. If you’re...
youtube.com
I tested @ollama versus @vllm_project for a project I was running. vllm running a full weight model was 4x faster in concurrent requests then ollama with a Q4. michaelcizmar.com/blog/2026/03...
vLLM beats Ollama in throughput by 4x at scale | Michael Cizmar
Benchmarking vLLM against Ollama for a multimodal knowledge graph extraction task.
michaelcizmar.com
Most #AI teams don't have an evaluation strategy. They have an evaluation event. Tested before launch. Shipped. Never looked back. Meanwhile 3 new models dropped this month, 2 cheaper, 1 probably better for the task. Here's a 5-step framework 2 fix that: michaelcizmar.com/blog/2026/03... #MLOps
A Concrete Evaluation Framework for LLM-Powered Pipelines | Michael Cizmar
How to stay confident in your model choices when the landscape changes every 90 days
michaelcizmar.com
The #EpsteinFiles aren't just a political story. They're a #KYC audit nobody asked for - and most PE and hedge fund risk frameworks failed it. #AI solutions do not scale easily and then the problem becomes the data... michaelcizmar.com/blog/2026/02...
The Epstein Files Are a Risk Management Case Study Nobody Asked For | Michael Cizmar
The Epstein Files exposed a KYC blind spot that no sanctions list can fix — here's what modern market intelligence infrastructure needs to do differently.
michaelcizmar.com
I'm already tired of hearing about the #hussleculture. It seems to project laziness and apathy. While running an award winning small business for 20 years, I tried to instill 1 thought: "If you are not exceptional, you are obsolete". #leadership #oneteam #culture www.linkedin.com/analytics/po...
LinkedIn Login, Sign in | LinkedIn
Login to LinkedIn to keep in touch with people you know, share ideas, and build your career.
linkedin.com
AI projects fail bcause they’r chartered poorly. I built #Purrview 2 prove the opposite...a tiny tool that detects when a cat enters frame & records it. 1) No POC. 2) No roadmap. 3) No “AI transformation.” 4) Use what works & it works.#AI isn’t failing.AI projects are. linkedin.com/pulse/purrvi...
Purrview: The Tiny AI Project That Worked—And Why Most Don’t
Most AI projects fail. Not because of missing technology.
linkedin.com
“Hey Michael, this event page is pure AI slop.” Fair enough, but dont judge a book by its cover. U, on the other hand, still have 1 hour to join 2 of the most experienced AI practitioners as they share notes from the field bfore it turns in2 actual slop: events.teams.microsoft.com/event/cc5517...
Microsoft Virtual Events Powered by Teams
Microsoft Virtual Events Powered by Teams
events.teams.microsoft.com
Simple is generally better than complex, I just published Auto-Generating Related Articles with @qdrant.bsky.social and Local Embeddings medium.com/p/auto-gener... #KNN
Auto-Generating Related Articles with QDrant and Local Embeddings
We are working on a new website at MC+A, and we’re always looking for ways to incorporate our tradecraft externally and we thought of a way…
medium.com
Why #PostgreSQL Search Isn’t Enough: A Case for Purpose-Built Retrieval Systems #VectorSearch #LTR #Elasticsearch @mcplusa mcplusa.com/why-postgres... #Insights
Why PostgreSQL Search Isn’t Enough: A Case for Purpose-Built Retrieval Systems
Postgres is a great database — but it’s not a search engine.Instacart’s recent move to consolidate search on Postgres highlights the risks: poor autocomplete, weak ranking, and frustrated users.If rel...
mcplusa.com
Checkout this image I made with #chatgpt5. The prompt was ‘generate me an image of me getting after it’
#MCP does not kill the univeral index, it really just extends it and like all things...it depends on the use case if universal versus #federatedsearch makes sense. mcplusa.com/why-federate...
Why Federated Search is Still Relevant Today
Glean’s recent blog post claims that federated search is on its way out. However, federated search is more relevant today than ever, and protocols like MCP are enhancing its effectiveness
mcplusa.com
Projects get stuck in the POC Production often because we do not describe the behavior or the work steams we aim to proof. "Is this an image of a cat?" - Asked no one. "Is this claim something we should further review" - Priceless. #LLM #BDD #TDD #Agents michaelcizmar.com/drive-requir...
Drive Requirements to Testing with BDD to Deliver AI
Successful AI Projects are Focused on Outcomes, Not Simply Outputs
michaelcizmar.com
Having the Title be : "Plastic Foodservice Film" and your LLM creating the synthentic queries is not the best method to judge your LLM's performance at finding Seran Rap or saran wrap. #LLM #AI #Relevancy michaelcizmar.com/judging-llm-...
Judging LLM Performance By Synthetic Data Is A Failing Approach — Part 1
Knock-offs are never as good as the real thing
michaelcizmar.com