[{"content":"","date":"12 April 2026","externalUrl":null,"permalink":"/tags/benchmarking/","section":"Tags","summary":"","title":"Benchmarking","type":"tags"},{"content":"","date":"12 April 2026","externalUrl":null,"permalink":"/blog/","section":"Blog","summary":"","title":"Blog","type":"blog"},{"content":"","date":"12 April 2026","externalUrl":null,"permalink":"/tags/cli/","section":"Tags","summary":"","title":"Cli","type":"tags"},{"content":"","date":"12 April 2026","externalUrl":null,"permalink":"/tags/coding-agents/","section":"Tags","summary":"","title":"Coding-Agents","type":"tags"},{"content":" Your browser does not support the audio element. Episode Notes # This is an example podcast episode page. Replace the audio source above with a real audio file or embed from a podcast platform.\nLinks # Resources mentioned in this episode ","date":"12 April 2026","externalUrl":null,"permalink":"/podcasts/example-episode/","section":"Podcasts","summary":"A placeholder episode to demonstrate the podcast section layout.","title":"Episode 0: Welcome to Yonk-Labs","type":"podcasts"},{"content":"","date":"12 April 2026","externalUrl":null,"permalink":"/tags/example/","section":"Tags","summary":"","title":"Example","type":"tags"},{"content":" Show Notes # This is an example video page. Replace the YouTube ID above with a real video.\nUse Hugo\u0026rsquo;s built-in youtube shortcode for YouTube embeds, or use a raw iframe for Vimeo and other platforms.\n","date":"12 April 2026","externalUrl":null,"permalink":"/videos/example-video/","section":"Videos","summary":"A placeholder video to demonstrate the video section layout.","title":"Example Video","type":"videos"},{"content":"","date":"12 April 2026","externalUrl":null,"permalink":"/tags/gallery/","section":"Tags","summary":"","title":"Gallery","type":"tags"},{"content":"","date":"12 April 2026","externalUrl":null,"permalink":"/tags/go/","section":"Tags","summary":"","title":"Go","type":"tags"},{"content":"This is the first post on the Yonk-Labs blog.\nWe started Yonk-Labs as a place to build open source tools and share what we learn. Expect technical deep dives, project announcements, and the occasional tutorial.\nWhat\u0026rsquo;s coming # Project showcases with technical breakdowns Tutorials and guides Lessons learned from building in the open Stay tuned.\n","date":"12 April 2026","externalUrl":null,"permalink":"/blog/hello-world/","section":"Blog","summary":"First post on the Yonk-Labs blog — what we’re building and why.","title":"Hello World","type":"blog"},{"content":"A minimal Go program to get started.\nThe Code # package main import \u0026#34;fmt\u0026#34; func main() { fmt.Println(\u0026#34;Hello from Yonk-Labs!\u0026#34;) } How to Run # go run main.go Full Source # See the complete example on GitHub.\n","date":"12 April 2026","externalUrl":null,"permalink":"/samples/example-sample/","section":"Sample Code","summary":"A simple Hello World example in Go.","title":"Hello World in Go","type":"samples"},{"content":"","date":"12 April 2026","externalUrl":null,"permalink":"/tags/hot-takes/","section":"Tags","summary":"","title":"Hot-Takes","type":"tags"},{"content":"","date":"12 April 2026","externalUrl":null,"permalink":"/tags/humor/","section":"Tags","summary":"","title":"Humor","type":"tags"},{"content":"","date":"12 April 2026","externalUrl":null,"permalink":"/humor/","section":"Humor \u0026 Yonk-isms","summary":"","title":"Humor \u0026 Yonk-isms","type":"humor"},{"content":"","date":"12 April 2026","externalUrl":null,"permalink":"/tags/json/","section":"Tags","summary":"","title":"Json","type":"tags"},{"content":"","date":"12 April 2026","externalUrl":null,"permalink":"/tags/llm/","section":"Tags","summary":"","title":"Llm","type":"tags"},{"content":"llm-top is a top-style terminal dashboard for monitoring LLM inference workloads running on NVIDIA DGX Spark (GB10). Get real-time visibility into GPU utilization, memory, processes, containers, and model health — all in one live-updating view.\nOverview # If you\u0026rsquo;re running vLLM, SGLang, NIM, or other LLM inference servers on a DGX Spark, llm-top gives you an at-a-glance view of:\nGPU stats: SM utilization, memory bandwidth, temperature, power, clock, and memory usage Host stats: CPU, RAM, core count GPU processes: PID, name, memory usage, and type (compute/graphics) Containers: CPU%, memory, network I/O, block I/O, PID counts Model servers: Port, health, request counts, KV cache usage, RPS, token throughput Tech Stack # Language: Python Target: NVIDIA DGX Spark (GB10) Links # GitHub Repository License: Apache 2.0 ","date":"12 April 2026","externalUrl":"https://github.com/TheYonk/llm-top","permalink":"/projects/llm-top/","section":"Projects","summary":"A top-like live terminal dashboard for monitoring LLM inference servers on NVIDIA DGX Spark.","title":"llm-top","type":"projects"},{"content":"","date":"12 April 2026","externalUrl":null,"permalink":"/tags/mcp/","section":"Tags","summary":"","title":"Mcp","type":"tags"},{"content":"","date":"12 April 2026","externalUrl":null,"permalink":"/tags/memes/","section":"Tags","summary":"","title":"Memes","type":"tags"},{"content":"","date":"12 April 2026","externalUrl":null,"permalink":"/tags/meta/","section":"Tags","summary":"","title":"Meta","type":"tags"},{"content":"","date":"12 April 2026","externalUrl":null,"permalink":"/tags/migration-testing/","section":"Tags","summary":"","title":"Migration-Testing","type":"tags"},{"content":"","date":"12 April 2026","externalUrl":null,"permalink":"/tags/mongodb/","section":"Tags","summary":"","title":"Mongodb","type":"tags"},{"content":"","date":"12 April 2026","externalUrl":null,"permalink":"/tags/monitoring/","section":"Tags","summary":"","title":"Monitoring","type":"tags"},{"content":"","date":"12 April 2026","externalUrl":null,"permalink":"/tags/mysql/","section":"Tags","summary":"","title":"Mysql","type":"tags"},{"content":"","date":"12 April 2026","externalUrl":null,"permalink":"/tags/nvidia/","section":"Tags","summary":"","title":"Nvidia","type":"tags"},{"content":"os-db-json-tester is a benchmarking and load-generation suite for comparing how MySQL, PostgreSQL, and MongoDB handle JSON workloads. It seeds a movie/actor dataset, runs a configurable website-style workload, and lets you probe query performance and concurrency behavior across all three databases with the same harness.\nOverview # Each database has different strengths when it comes to JSON: PostgreSQL has jsonb with GIN indexes, MySQL has JSON columns and generated-column indexes, and MongoDB is document-native. os-db-json-tester gives you a single tool to stress-test them all with a realistic workload (loading actors, titles, movies, and directors; running mixed read/write traffic).\nTech Stack # Language: Python Targets: PostgreSQL, MySQL, MongoDB Packaging: Docker container (PostgreSQL target) Quick Start # Local (macOS):\nbrew install mysql pkg-config pip install -r requirement.txt python bench/app_controller.py -f ./bench/app_config/xxx.json Docker (PostgreSQL):\ndocker build -t os-db-json-tester . docker run -it --env-file .envfile os-db-json-tester The .envfile takes HOSTNAME, USERNAME, PASSWORD, and DATABASE — start_bench.sh uses these to generate the runtime config.\nLinks # GitHub Repository License: MIT ","date":"12 April 2026","externalUrl":"https://github.com/TheYonk/os-db-json-tester","permalink":"/projects/os-db-json-tester/","section":"Projects","summary":"Scripts to test and benchmark JSON functionality across MySQL, PostgreSQL, and MongoDB — and generate simulated load.","title":"os-db-json-tester","type":"projects"},{"content":"pg-retest captures production PostgreSQL traffic and replays it against test targets so you can validate configuration changes, version upgrades, server migrations, and capacity plans before they hit production. Think of it as Oracle RAT for PostgreSQL — with modern tooling and a demo environment you can spin up in one command.\nWhy pg-retest? # Pre-migration validation — Replay production traffic against your new datacenter, hardware, or cloud target before cutting over. Version \u0026amp; patch testing — Upgrading PostgreSQL 15 → 16? Replay your exact workload and catch regressions before they ship. Configuration benchmarking — Changed shared_buffers or work_mem? Compare before and after with real queries, not synthetic benchmarks. Cloud provider evaluation — RDS vs. Aurora vs. AlloyDB vs. self-hosted — replay identical traffic against each. Capacity planning — Scale workloads 2x, 5x, 10x to find where things break before Black Friday does. CI/CD regression gates — Automated pass/fail on every schema migration or config change. Cross-database migration — Moving from MySQL to PostgreSQL? Capture, transform SQL, and validate on PG. AI-assisted optimization — LLM-powered tuning recommendations, validated against your real workload with automatic rollback on regression. Accuracy # pg-retest is a workload simulation tool, not a replication system. Replay produces a high-fidelity approximation of production traffic: 93-96% accuracy for write workloads with --id-mode=full, near-100% for read-only. The 4-7% error on writes comes from concurrent session sequence ordering — the same fundamental limitation Oracle RAT documents as \u0026ldquo;replay divergence.\u0026rdquo;\nUse pg-retest to answer \u0026ldquo;will my workload perform the same on the new target?\u0026rdquo; — not \u0026ldquo;will my data be byte-identical?\u0026rdquo;\nTech Stack # Language: Rust Target: PostgreSQL 12+ Quality bar: 323 tests, zero clippy warnings Quick Start # git clone https://github.com/pg-retest/pg-retest.git cd pg-retest docker compose up --build # Open http://localhost:8080 and click \u0026#34;Demo\u0026#34; The demo ships two PostgreSQL 16 instances seeded with a 94k-row e-commerce dataset and a pre-built workload (357 queries, 8 concurrent sessions). The Demo page walks you through inspect → replay → compare → scale → AI tuning.\nLinks # GitHub Repository License: Apache 2.0 ","date":"12 April 2026","externalUrl":"https://github.com/pg-retest/pg-retest","permalink":"/projects/pg-retest/","section":"Projects","summary":"Capture, replay, and compare PostgreSQL workloads. Validate config changes, migrations, and capacity plans with confidence.","title":"pg-retest","type":"projects"},{"content":"","date":"12 April 2026","externalUrl":null,"permalink":"/tags/pgvector/","section":"Tags","summary":"","title":"Pgvector","type":"tags"},{"content":"","date":"12 April 2026","externalUrl":null,"permalink":"/podcasts/","section":"Podcasts","summary":"","title":"Podcasts","type":"podcasts"},{"content":"","date":"12 April 2026","externalUrl":null,"permalink":"/tags/postgres/","section":"Tags","summary":"","title":"Postgres","type":"tags"},{"content":"","date":"12 April 2026","externalUrl":null,"permalink":"/tags/postgresql/","section":"Tags","summary":"","title":"Postgresql","type":"tags"},{"content":"","date":"12 April 2026","externalUrl":null,"permalink":"/projects/","section":"Projects","summary":"","title":"Projects","type":"projects"},{"content":"","date":"12 April 2026","externalUrl":null,"permalink":"/tags/python/","section":"Tags","summary":"","title":"Python","type":"tags"},{"content":"","date":"12 April 2026","externalUrl":null,"permalink":"/tags/rag/","section":"Tags","summary":"","title":"Rag","type":"tags"},{"content":"RoboMonkey MCP is a local-first Model Context Protocol server that indexes code and documentation into PostgreSQL with pgvector, then provides hybrid retrieval — vector search + full-text search + tags — to any MCP-compatible coding client (Claude Desktop, Claude Code, etc.).\nOrigin Story # \u0026ldquo;100% unapologetically vibe coded with the assistance of Claude. The idea came from working on a series of video games over winter break and running into problems with coding agents losing info about files and not understanding the relationships between files.\u0026rdquo;\nThat\u0026rsquo;s the honest README. RoboMonkey exists because coding agents need persistent, structured memory of your codebase — not a fresh grep on every turn.\nWhat It Does # Indexes your codebase — analyzes code structure, symbols, and relationships Hybrid retrieval — combines dense vector embeddings, BM25 full-text search, and tag filtering Background daemon — watches files, processes changes incrementally, keeps the index fresh MCP integration — exposes tools to Claude Desktop via a generated .mcp.json Pluggable embeddings — Ollama, vLLM, or OpenAI Tech Stack # Language: Python 3.11+ Storage: PostgreSQL with pgvector Runtime: Docker Compose for Postgres, background Python daemon for watching/indexing Embeddings: Ollama / vLLM / OpenAI Interface: MCP server Quick Start # ./quick_start.sh The script creates a venv, starts Postgres+pgvector in Docker, configures your embedding and LLM providers interactively, indexes a repo of your choice, starts the daemon, and writes .mcp.json for Claude Desktop. Takedown is a single ./quick_teardown.sh.\nLinks # GitHub Repository License: Apache 2.0 ","date":"12 April 2026","externalUrl":"https://github.com/TheYonk/yonk-robo-codemonkey","permalink":"/projects/yonk-robo-codemonkey/","section":"Projects","summary":"Local-first MCP server that indexes code and docs into Postgres + pgvector for hybrid retrieval by LLM coding clients.","title":"RoboMonkey MCP (yonk-robo-codemonkey)","type":"projects"},{"content":"","date":"12 April 2026","externalUrl":null,"permalink":"/tags/rust/","section":"Tags","summary":"","title":"Rust","type":"tags"},{"content":"","date":"12 April 2026","externalUrl":null,"permalink":"/samples/","section":"Sample Code","summary":"","title":"Sample Code","type":"samples"},{"content":"","date":"12 April 2026","externalUrl":null,"permalink":"/tags/","section":"Tags","summary":"","title":"Tags","type":"tags"},{"content":"A picture is worth a thousand words, and a good meme is worth about forty technical blog posts. Here are some of the ones I\u0026rsquo;ve made or posted in the wild, usually right after sitting through a vendor pitch that made me lose faith in humanity.\nClick any image to see it full size. Save them, share them, use them in your own decks — I\u0026rsquo;m not precious about it.\nDo You Feel Lucky?\nFor every time somebody asked me to deploy a schema change on a Friday afternoon to production without a backup. Go ahead. Make my day. Welcome to NoSQL Land\nThe happiest place on earth, until you need a join. Then it\u0026#39;s just a regular database with more steps. Arcade Tokens\nRemember when tokens meant three lives on Street Fighter II? Now they mean your CFO is having a very hard Tuesday. \u0026#34;It\u0026#39;s Just Tokens\u0026#34;\nHow every AI vendor explains their pricing, moments before you get the first invoice. NL2SQL: WTF Edition\nWhen the demo hits production and the LLM starts writing queries that would make a DBA file for PTO. More coming as I make them. If a database company says something dumb at a keynote, it will probably show up here within 48 hours.\n","date":"12 April 2026","externalUrl":null,"permalink":"/humor/gallery/","section":"Humor \u0026 Yonk-isms","summary":"Memes, cartoons, and visual shenanigans I’ve made (or inflicted on people) over the years.","title":"The Gallery","type":"humor"},{"content":"After 20+ years of ranting at conferences, recording podcasts, and explaining databases to people who\u0026rsquo;d rather be doing literally anything else, you develop a vocabulary. This is mine. Some of it is useful. Some of it is me being a smart-ass. All of it has shown up in a meeting where somebody stared at me and said \u0026ldquo;wait, is that a bit, or are you serious?\u0026rdquo;\nThe answer is always yes.\nThe hot takes I will not shut up about # \u0026ldquo;Database design is a lost art.\u0026rdquo; # This is the one that gets the most eye-rolls and the most slow-nodding-in-agreement, usually from the same people. We stopped teaching it. We stopped caring about it. And now half the \u0026ldquo;AI problems\u0026rdquo; I see are actually \u0026ldquo;somebody didn\u0026rsquo;t normalize their schema\u0026rdquo; problems wearing a trench coat.\n\u0026ldquo;In AI Land, the hard problem is always data engineering, not database selection.\u0026rdquo; # Your RAG pipeline is not failing because you picked Postgres over Milvus. It\u0026rsquo;s failing because your chunking strategy is garbage and your embedding model has no idea what domain you\u0026rsquo;re in. I will repeat this until I lose my voice.\n\u0026ldquo;Right-sized AI.\u0026rdquo; # Not every problem needs a 400-billion-parameter model. A 7B model running locally with a good retrieval layer and honest evaluation will beat GPT-4 on your specific task 60% of the time, at 5% of the cost, with none of the data leaving your building. You want fit, not flex.\n\u0026ldquo;Small things in application design have a big impact on database performance.\u0026rdquo; # I\u0026rsquo;ve spent entire consulting engagements fixing a single bad ORM pattern. One SELECT * in a loop can take down a cluster. One missing index can cost a company millions. The people who design the app are making performance decisions for the DBAs whether they know it or not.\n\u0026ldquo;Show me the methodology, or your benchmark is marketing.\u0026rdquo; # Any benchmark without workload, concurrency, hardware, dataset size, and version is not a benchmark. It\u0026rsquo;s a press release. If your comparison graph starts the Y-axis at 80% to make a 2% difference look like a victory, I am already mad at you.\nThe Oreo Analogy # I came up with this one to explain where PostgreSQL sits in the modern AI stack, and somehow it stuck. The story goes:\nYour AI stack is an Oreo. The top cookie is the model providers — OpenAI, Anthropic, Google, the usual suspects. The bottom cookie is the agent frameworks — LangChain, LlamaIndex, AutoGen, whatever\u0026rsquo;s trendy this week.\nBut here\u0026rsquo;s the thing. The Oreo isn\u0026rsquo;t the cookies. The Oreo is the double stuff in the middle. The data layer. The governance layer. The retrieval layer. The thing that actually knows what your enterprise data is and who\u0026rsquo;s allowed to see it.\nThat\u0026rsquo;s the database. That\u0026rsquo;s PostgreSQL. And if you don\u0026rsquo;t get the middle right, no amount of cookie is going to save your cookie.\nPeople have stolen this one. I don\u0026rsquo;t mind. It works.\nThe Verbal Tics # Things I apparently say a lot, based on transcripts I\u0026rsquo;ve been subjected to:\n\u0026ldquo;Look, \u0026hellip;\u0026rdquo; — how I start any sentence where I\u0026rsquo;m about to disagree with you politely \u0026ldquo;Here\u0026rsquo;s the deal\u0026hellip;\u0026rdquo; — how I start any sentence where the disagreement is no longer going to be polite \u0026ldquo;So what does that actually mean?\u0026rdquo; — my defense mechanism against hand-wavy marketing \u0026ldquo;I have thoughts.\u0026rdquo; — said right before I have more than thoughts, I have a 40-minute rant \u0026ldquo;Sorry, not sorry.\u0026rdquo; — deployed after any statement a vendor is going to be mad about \u0026quot;(don\u0026rsquo;t @ me)\u0026quot; — parenthetical, always in writing, usually right after I\u0026rsquo;ve said something everybody is going to @ me about \u0026ldquo;The HOSS\u0026rdquo; — referring to myself in the third person, tongue firmly in cheek, because if I\u0026rsquo;m going to have an absurd title I\u0026rsquo;m going to lean into it The Recurring Bits # Wrestling analogies for everything # I build wrestling sims in my free time. Occupational hazard: I now explain every technical concept via wrestling. \u0026ldquo;Postgres and MongoDB aren\u0026rsquo;t enemies, they\u0026rsquo;re a tag team.\u0026rdquo; \u0026ldquo;Your DBaaS provider is your promoter and you\u0026rsquo;re the talent — know the difference.\u0026rdquo; \u0026ldquo;Database replication is a face-heel turn in slow motion.\u0026rdquo;\nIf this makes no sense to you, congratulations, you are a healthy person. If it does make sense to you, we need to have a beer.\nCooking metaphors for team building # I cook for my teams. Sometimes over video calls, which is exactly as weird as it sounds. I will 100% explain customer success strategy by making you watch me make a risotto. The metaphor is: you cannot rush it, you cannot skip the stirring, and if you walk away from the stove you will deserve what you get. Database operations are the same. Team building is the same. Most things are the same.\nReferring to past selves in the third person # \u0026ldquo;There was a version of me in 2011 who thought MongoDB was going to eat everything. That guy was young and wrong. I bought him a beer and told him I was sorry.\u0026rdquo;\nThe fact that I picked my own title # HOSS stands for Head of Open Source Strategy. I picked it. When I got to Percona the second time and they asked what my title should be, I said \u0026ldquo;The HOSS\u0026rdquo; and everybody laughed and I said \u0026ldquo;no, seriously\u0026rdquo; and now it\u0026rsquo;s on multiple conference badges.\nThe Retired Bits # These used to be in rotation. They have been lovingly put out to pasture:\n\u0026ldquo;MongoDB is web scale.\u0026rdquo; — the joke was funny in 2010. Let it go. \u0026ldquo;NoSQL is just SQL with fewer letters.\u0026rdquo; — still accurate, but I don\u0026rsquo;t want to start another war. \u0026ldquo;Kubernetes is a distributed system for distributing the problem of running a distributed system.\u0026rdquo; — retired because it\u0026rsquo;s more depressing than funny now. If I say something in a podcast and you want me to add it to the list, let me know. The canon expands.\n","date":"12 April 2026","externalUrl":null,"permalink":"/humor/yonk-isms/","section":"Humor \u0026 Yonk-isms","summary":"A running list of things I say too often, beliefs I will not shut up about, and the recurring bits that turned into actual brand.","title":"The Yonk-isms","type":"humor"},{"content":"","date":"12 April 2026","externalUrl":null,"permalink":"/videos/","section":"Videos","summary":"","title":"Videos","type":"videos"},{"content":"","date":"12 April 2026","externalUrl":null,"permalink":"/tags/workload-replay/","section":"Tags","summary":"","title":"Workload-Replay","type":"tags"},{"content":"","date":"12 April 2026","externalUrl":null,"permalink":"/tags/yonk-isms/","section":"Tags","summary":"","title":"Yonk-Isms","type":"tags"},{"content":"Welcome to Yonk-Labs — we build open source tools and share what we learn along the way.\nExplore our blog for technical write-ups, watch our videos, listen to our podcasts, browse sample code, or check out our projects.\n","date":"12 April 2026","externalUrl":null,"permalink":"/","section":"Yonk-Labs","summary":"Welcome to Yonk-Labs — we build open source tools and share what we learn along the way.\nExplore our blog for technical write-ups, watch our videos, listen to our podcasts, browse sample code, or check out our projects.\n","title":"Yonk-Labs","type":"page"},{"content":"Hi. I\u0026rsquo;m Matt Yonkovit. Most people who\u0026rsquo;ve been around open source databases for a while know me as The Yonk. Or The HOSS, from the years I carried around the title Head of Open Source Strategy. I picked that one myself. I regret nothing.\nThis is where I park the stuff I build on nights and weekends — and increasingly, the stuff I build during the day that refuses to stay in a single company repo. Some of it is useful. Some of it is me arguing with myself in code. Occasionally those are the same thing.\nWhat I\u0026rsquo;m actually into # I like databases. Specifically, I like what happens inside them when nobody\u0026rsquo;s watching — buffer pools doing clever things, query planners making bad decisions, replication quietly falling over at 2am. I\u0026rsquo;ve spent 20+ years chasing that stuff, and somewhere along the way I ended up on the AI side of the fence too. Not because AI is shiny. Because the data problems behind AI — chunking, embeddings, hybrid retrieval, air-gapped inference on locked-down hardware — turned out to be the exact same flavor of problem I\u0026rsquo;ve been working on my whole career. Just with more GPUs and a fresh coat of hype.\nRight now I\u0026rsquo;m at EnterpriseDB, where I get paid to think about how PostgreSQL and AI should meet. In practice, that means a lot of Kubernetes-native operations, a lot of time on NVIDIA DGX hardware, a lot of debugging weird inference bugs, and a lot of arguing about governance. Agentic AI with JWT-scoped purpose controls and Postgres RLS as a hard policy layer. Right-sized models that don\u0026rsquo;t require you to ship your data off-prem. Inference pipelines in Rust. That kind of thing.\nBefore that, I spent a year or so running open source strategy at Scarf, helping maintainers actually see who was using their packages. Before that, I was at Percona for the better part of a decade, in roles that kept shape-shifting — Principal Architect, VP of Global Services, Chief Experience Officer, and eventually the HOSS. While I was there I worked extensively on setting the groundwork for Percona\u0026rsquo;s PMM and QAN observability tools, helped launch the Percona PostgreSQL and MongoDB Server product lines, built out customer success and delivery orgs, and generally spent a lot of time trying to get the engineering side and the business side to talk to each other. And before that I was contracting into MySQL AB / Sun Microsystems on InnoDB internals, which is where I learned that database performance is 10% clever algorithms and 90% \u0026ldquo;what is this one pathological query actually doing to the buffer pool.\u0026rdquo;\nI\u0026rsquo;ve run P\u0026amp;Ls, built DevRel teams, sat in enough boardrooms to know what not to say in them, and shipped a lot of open source. I\u0026rsquo;m a true believer and the guy who has had to sign the paychecks. People think those are in tension. They\u0026rsquo;re not. When community and business stop talking to each other is when projects die, and I have watched that movie more times than I would like.\nI also founded the Open Source Business Community — a public resource for metrics, guides, and podcasts about how to run an open source company without setting it on fire.\nThe podcasts and the videos # Two shows you might know me from:\nPercona\u0026rsquo;s HOSS Talks FOSS — the show where I sit down with developers, DBAs, CEOs, and community maintainers from all over the open source database world. MySQL, PostgreSQL, MongoDB, MariaDB, Cassandra, the whole menagerie. Guests have come from DataStax, Google, AWS, and basically every database company you can name. Video on YouTube, audio everywhere podcasts live. The Hacking Open Source Business Podcast — co-hosted with Avi Press of Scarf. This one is less about the code and more about the business of open source. Licensing drama, VC incentives, commercialization strategies, what actually works, what quietly doesn\u0026rsquo;t. If you run or work at an open source company and you\u0026rsquo;ve ever wondered why the numbers don\u0026rsquo;t add up, this show is for you. I\u0026rsquo;ve also done a lot of conference talks over the years — FOSDEM (usually on data engineering and RAG accuracy these days), PGConf, FOSSY, Open Source Summit, All Things Open, and obviously Percona Live. If you\u0026rsquo;ve sat through one of my sessions, I apologize and I hope the jokes landed.\nWhat I\u0026rsquo;ll be putting in the videos and podcasts sections of this site is a little different. More personal. Experiments, demos, unfiltered rants, and the kind of stuff that doesn\u0026rsquo;t fit cleanly into a corporate feed. Think of it as the directors-cut-with-commentary version of whatever I\u0026rsquo;m obsessing over that week.\nWhat I\u0026rsquo;ll rant about if you buy me a beer # Data engineering is the hard part of AI. Not the model. Not the prompt. The pipeline. The chunking. The embedding strategy. The filtering on retrieval. I\u0026rsquo;ve been benchmarking this stuff for months and presenting the results on conference stages, and I have thoughts — most of them are going to make a few vendors uncomfortable. Sorry, not sorry.\nRight-sized AI is a thing. Not everything needs a 400-billion-parameter model. A lot of \u0026ldquo;AI problems\u0026rdquo; are solved better by a purpose-built model under 10B parameters running locally, plus a database that actually knows what it\u0026rsquo;s doing. PostgreSQL + pgvector + a small embedding model will eat 80% of the use cases people are paying OpenAI rates for. I\u0026rsquo;ll die on this hill.\nAir-gapped AI is real and it\u0026rsquo;s coming for you. Regulated enterprises, government, healthcare — they\u0026rsquo;re going to run LLM inference on-prem or nowhere, and the tooling for that is not where it needs to be yet. Half of my current work is fixing that.\nOpen source communities need business models, and business models need communities. The companies that forget this — on either side — become cautionary tales. I\u0026rsquo;ve watched it happen at places you\u0026rsquo;ve heard of, and it\u0026rsquo;s always the same pattern.\nHonest benchmarks over marketing benchmarks. If your graph starts at 80% of the Y-axis to make a 2% improvement look like a victory, I am already mad at you.\nDatabase design is a lost art. There, I said it.\nThe projects here # Everything in the projects section is something I actually use or have used. A few highlights:\npg-retest — Oracle RAT, but for PostgreSQL, in Rust, with a demo you can spin up in one command. Capture production traffic, replay it against a new target, know before you cut over. llm-top — top but for LLM inference servers on NVIDIA DGX Spark. Because I got tired of opening four terminals to watch one GPU. RoboMonkey MCP — A local-first MCP server that indexes code and docs into Postgres with pgvector so coding agents stop losing the plot on long sessions. Yes, it is 100% unapologetically vibe-coded. That\u0026rsquo;s the origin story and I\u0026rsquo;m keeping it. os-db-json-tester — Benchmark harness for JSON workloads across MySQL, PostgreSQL, and MongoDB. Exists because every vendor\u0026rsquo;s JSON benchmarks are, how do I put this politely, optimistic. Things that are not in my job description but I do anyway # Outside the database world, I build simulation games nobody asked for. A 44K-line wrestling booking simulator with feud heat tracking and tag team chemistry decay. A 53K-line meta-tool for building management sims. A multi-agent system for writing novels. I cook for my teams, sometimes over video calls, which is weirder than it sounds. I\u0026rsquo;m a genuine wrestling nerd (the real kind — I will absolutely argue the business of kayfabe with you).\nNone of this is strategy. It\u0026rsquo;s just what happens when you let a curious person have too much unstructured time and a working laptop.\nWhy this site exists # Three reasons, because I list things in threes:\nI build a lot of stuff and needed somewhere to put it that wasn\u0026rsquo;t twelve abandoned repos and a LinkedIn post. The blog, videos, and podcast sections are for the long-form rants. I\u0026rsquo;d rather write these down than lose the argument to whoever yelled back at me on Twitter. It\u0026rsquo;s an excuse to dogfood my own \u0026ldquo;small static site, big ideas\u0026rdquo; belief. No JavaScript framework du jour. No analytics trying to follow you around. Just markdown, Hugo, and GitHub Pages. Find me # GitHub (Yonk-Labs org): github.com/yonk-labs — where the projects on this site live GitHub (personal): github.com/TheYonk — the weirder personal stuff, including the wrestling sim LinkedIn: linkedin.com/in/myonk Open Source Business Community: opensourcebusiness.community — the public resource I founded for running open source companies without setting them on fire HOSS Talks FOSS: percona.podbean.com Hacking Open Source Business Podcast: podcast.hosbp.com If you want to argue, build something together, or just tell me I\u0026rsquo;m wrong about MongoDB, any of the above will reach me. Or find me at any database conference where there\u0026rsquo;s good coffee and someone willing to listen.\nWelcome to Yonk-Labs. Pull up a chair.\n","externalUrl":null,"permalink":"/about/","section":"Yonk-Labs","summary":"Hi. I’m Matt Yonkovit. Most people who’ve been around open source databases for a while know me as The Yonk. Or The HOSS, from the years I carried around the title Head of Open Source Strategy. I picked that one myself. I regret nothing.\nThis is where I park the stuff I build on nights and weekends — and increasingly, the stuff I build during the day that refuses to stay in a single company repo. Some of it is useful. Some of it is me arguing with myself in code. Occasionally those are the same thing.\n","title":"About","type":"page"},{"content":"","externalUrl":null,"permalink":"/categories/","section":"Categories","summary":"","title":"Categories","type":"categories"}]