Publication

Practical LLM Systems

@practical-llm-systems

articles: 26
total claps: 374
top claps: 85
avg read: 8.0 min
paywalled: 8
active: 2026-05-14 → 2026-07-15

Wiki topics

AI · AI · General · 26 💻 · Programming · 21 LLM · Large Language Models · 16 AGT · AI Agents · 7 ML · Machine Learning · 7 EDU · Education & Learning · 6 OPS · LLMOps & Inference · 5 🏛️ · Architecture · 4

Top tags

#technology ·26 #artificial-intelligence ·26 #programming ·21 #software-engineering ·19 #llm ·10 #machine-learning ·6 #software-development ·5 #ai ·3 #software-architecture ·2 #data-science ·2

AI TCH

RO Rost Glukhov · Practical LLM Systems

I Tested MTP Speculative Decoding on Two Qwen Models — One Was a Trap

You enable speculative decoding, see generation speed jump 67%, and think you’ve cracked fast local inference. Then you check your context…

May 21, 2026 · 6 min read · 85 claps #technology #programming #artificial-intelligence

I Tested MTP Speculative Decoding on Two Qwen Models — One Was a Trap

AI TCH SPT

RO Rost Glukhov · Practical LLM Systems

Hermes Agent: Headless Server + Remote Desktop Setup

Running Hermes Agent on a headless server while connecting from a desktop client on another machine requires two server processes and a…

Jul 13, 2026 · 4 min read · 62 claps paywalled #technology #programming #artificial-intelligence

Hermes Agent: Headless Server + Remote Desktop Setup

AI HUM TCH SPT

RO Rost Glukhov · Practical LLM Systems

Stop Exposing Ollama to the Internet — Here’s the Safe Way

I watched someone forward port 11434 on their router one day. The next week, their GPU was running at 100% with unknown traffic.

May 27, 2026 · 5 min read · 62 claps #technology #programming #artificial-intelligence

Stop Exposing Ollama to the Internet — Here’s the Safe Way

AI HUM

RO Rost Glukhov · Practical LLM Systems

A2A vs MCP: Do AI Agents Really Need Both Protocols?

MCP gives agents tools. A2A gives agents peers.

Jun 25, 2026 · 11 min read · 50 claps paywalled #technology #artificial-intelligence #software-engineering

A2A vs MCP: Do AI Agents Really Need Both Protocols?

AI SCI LIF

RO Rost Glukhov · Practical LLM Systems

Your Notes Are Not a Second Brain — Here’s What Is

You have 47 markdown files, three Notion workspaces, and a folder called “Research” that hasn’t been opened since last March. Sound…

May 15, 2026 · 6 min read · 31 claps #technology #artificial-intelligence #knowledge

Your Notes Are Not a Second Brain — Here’s What Is

AI MDA TCH SPT

RO Rost Glukhov · Practical LLM Systems

A2A Streaming and Async Tasks for Long-Running Agent Workflows

Most AI agent demos still behave like chat completions with extra steps: you send a prompt, wait a few seconds, and get an answer back in…

Jul 15, 2026 · 11 min read · 20 claps paywalled #technology #programming #artificial-intelligence

A2A Streaming and Async Tasks for Long-Running Agent Workflows

AI ART TCH

RO Rost Glukhov · Practical LLM Systems

Decision Records for AI-Driven Software Development

Keep intent close to the code.

Jul 06, 2026 · 15 min read · 19 claps paywalled #technology #programming #artificial-intelligence

Decision Records for AI-Driven Software Development

AI TCH

RO Rost Glukhov · Practical LLM Systems

A2A and MCP Agent Security: Identity, Delegation, and Audit Trails

Prompt injection gets most of the security attention in LLM systems, and it deserves attention, but it is not the whole problem once agents…

Jul 12, 2026 · 9 min read · 17 claps paywalled #technology #programming #artificial-intelligence

A2A and MCP Agent Security: Identity, Delegation, and Audit Trails

AI TCH

RO Rost Glukhov · Practical LLM Systems

Your GPU Is Full. llama.cpp Has No “Unload All” Button. Here’s the Fix.

You load three models into llama-server router mode. Then you want them gone.

May 17, 2026 · 5 min read · 12 claps #technology #programming #artificial-intelligence

Your GPU Is Full. llama.cpp Has No “Unload All” Button. Here’s the Fix.

AI TCH

RO Rost Glukhov · Practical LLM Systems

Most Hermes Skills Are Broken — Here’s Why

You write a skill. It works in your terminal. You push it to your team. Three people never see it. Two more invoke it but get wrong…

May 18, 2026 · 6 min read · 8 claps #programming #technology #artificial-intelligence

Most Hermes Skills Are Broken — Here’s Why

AI HUM TCH

RO Rost Glukhov · Practical LLM Systems

Qwen 3.6 27B and 35B MTP vs Standard on 16GB GPU

MTP vs standard decoding on RTX 4080 — real benchmarks

Jun 22, 2026 · 7 min read · 7 claps #technology #programming #artificial-intelligence

Qwen 3.6 27B and 35B MTP vs Standard on 16GB GPU

AI TCH

RO Rost Glukhov · Practical LLM Systems

What Is Spec-Driven Development? The Spec as Source of Truth

Spec-Driven Development is one of those ideas that software engineers have reached for before and then set aside when the effort stopped…

Jul 07, 2026 · 9 min read · 1 claps paywalled #technology #programming #artificial-intelligence

What Is Spec-Driven Development? The Spec as Source of Truth

AI ART GEN TCH

RO Rost Glukhov · Practical LLM Systems

Most AI Assistant Memory Systems more complex — Here’s Why

Your AI assistant remembers nothing between sessions. You add a vector database. It starts remembering things — wrong things, at the wrong…

Jun 12, 2026 · 6 min read · 0 claps #technology #programming #artificial-intelligence

Most AI Assistant Memory Systems more complex — Here’s Why

AI ART TCH

RO Rost Glukhov · Practical LLM Systems

Polling Agents in AI Assistants: 11 Implementation Patterns

Reliable polling patterns for AI agents.

Jun 27, 2026 · 16 min read · 0 claps #artificial-intelligence #technology #programming

Polling Agents in AI Assistants: 11 Implementation Patterns

AI TCH

RO Rost Glukhov · Practical LLM Systems

Oh My Opencode Review: Honest Results, Billing Risks, and When It’s Worth It

What actually happens when you run Ultrawork.

Jun 17, 2026 · 8 min read · 0 claps #software-engineering #technology #programming

Oh My Opencode Review: Honest Results, Billing Risks, and When It’s Worth It

AI TCH SCI

RO Rost Glukhov · Practical LLM Systems

The Four Knowledge Systems Almost Nobody Gets Right

You have a wiki full of half-formed scratch notes, a RAG system confidently surfacing outdated information, and an AI agent that…

May 15, 2026 · 6 min read · 0 claps #technology #programming #artificial-intelligence

The Four Knowledge Systems Almost Nobody Gets Right

AI ECO TCH

RO Rost Glukhov · Practical LLM Systems

Cost Optimization for LLM Systems: Where the Money Actually Goes

Spend tokens where they actually matter.

Jun 20, 2026 · 4 min read · 0 claps #programming #artificial-intelligence #technology

Cost Optimization for LLM Systems: Where the Money Actually Goes

AI HUM TCH SCI

RO Rost Glukhov · Practical LLM Systems

Speculative Decoding: 20–50% Faster LLM Inference

Faster LLM inference without quality loss — a practical guide

Jul 05, 2026 · 16 min read · 0 claps paywalled #technology #programming #artificial-intelligence

Speculative Decoding: 20–50% Faster LLM Inference

AI TCH

RO Rost Glukhov · Practical LLM Systems

LLM Structured Output Validation in Python

Stop parsing vibes. Validate contracts.

May 14, 2026 · 13 min read · 0 claps #technology #programming #artificial-intelligence

LLM Structured Output Validation in Python

AI TCH

RO Rost Glukhov · Practical LLM Systems

Model Routing: Stop Using One Model for Everything

The right model for the right task.

Jun 19, 2026 · 4 min read · 0 claps #programming #artificial-intelligence #technology

Model Routing: Stop Using One Model for Everything

AI HUM LIF TCH

RO Rost Glukhov · Practical LLM Systems

Google A2A Protocol in 2026: Adoption, Hype, and Reality

A2A is not dead. It is just not universal.

Jun 29, 2026 · 16 min read · 0 claps #technology #programming #artificial-intelligence

Google A2A Protocol in 2026: Adoption, Hype, and Reality

AI LIF

RO Rost Glukhov · Practical LLM Systems

Most AI Knowledge Workflows Are Broken — Here’s Why

You feed your entire knowledge base into an LLM and get back something that sounds confident, reads smoothly, and is completely useless…

Jul 04, 2026 · 5 min read · 0 claps paywalled #technology #artificial-intelligence #software-engineering

Most AI Knowledge Workflows Are Broken — Here’s Why

AI HUM LIF SOC TCH

RO Rost Glukhov · Practical LLM Systems

GitHub Stars Are a Vanity Metric. Here’s the Real Adoption Data for AI Agents in 2026

On May 10, 2026, Hermes Agent surpassed OpenClaw as the most-used open-source AI agent in the world — by daily token volume on OpenRouter…

May 24, 2026 · 4 min read · 0 claps #artificial-intelligence #machine-learning #software-development

GitHub Stars Are a Vanity Metric. Here’s the Real Adoption Data for AI Agents in 2026

AI ART TCH DSN

RO Rost Glukhov · Practical LLM Systems

Why One Model Isn’t Enough: Multi-Model LLM Architecture Explained

You started with one model. It worked for your prototype. Then you added users, then edge cases, then a task the model flat-out refused to…

Jun 16, 2026 · 5 min read · 0 claps #programming #llm #artificial-intelligence

Why One Model Isn’t Enough: Multi-Model LLM Architecture Explained

AI TCH

RO Rost Glukhov · Practical LLM Systems

I Built a Local AI Search Engine — Here’s What Actually Works

I wanted an AI search engine that didn’t send my queries to a cloud provider. What I found instead was a networking nightmare, a confusing…

May 20, 2026 · 4 min read · 0 claps #technology #artificial-intelligence #llm