Machine Learning Mastery

But cutting your runtime token burn is just the first problem.

With the vocabulary and the failure modes in place, here's the build.

Day 100 in production isn't really about chunking strategies anymore.

This chapter is divided into eight parts; they are: • Metrics for LLM Inference • Measuring a Single Request • Warmup and Synchronization • Measuring GPU Work with CUDA Events • Measuring Memory Usage • Measuring Concurrent Requests • Multiple GPUs and Multiple Machines • Cost per Token The most common inference metrics are: • Latency: How long a request takes from start to finish.

In this article, you will learn how static, dynamic, and continuous batching work in LLM inference, and why the differences between them matter at production...

This chapter is divided into nine parts; they are: • Reading Logits from a Model • Greedy Decoding • Temperature Sampling • Top-$k$ Sampling • Nucleus Sampling • Repetition Penalties • Beam Search • Stop Conditions • Structured Output Constraints The model returns a vector of logits for every position in the input sequence.

This chapter is divided into four parts; they are: • Autoregressive Generation • Prefill and Decode • A Simple KV Cache • Memory Usage of the KV Cache A decoder-only transformer model predicts the next token from the tokens that came before it.

In this article, you will learn the seven architectural components that separate a production-grade agentic AI system from a demo script, and how each one...

In this article, you will learn how Ollama, LM Studio, and llama.cpp differ across the dimensions that matter most to practitioners, and how to choose...

This article introduces five architectural patterns for managing persistent memory and state in AI agents: In-Context Working Buffer, Execution Checkpointing, Semantic Memory, Episodic Event Logs, and Multi-Scope Segregation. These patterns address short-term execution, fault tolerance, cross-session knowledge, historical reflection, and multi-tenant data privacy respectively.

This article explores two paradigms for managing agent state—stateless and stateful—analyzing their scalability tradeoffs and illustrating implementations with Groq API code examples. Stateless agents are easy to scale horizontally but require the client to resend full conversation history, while stateful agents manage memory via a database but add architectural complexity.

Loop engineering is the practice of designing self-running cycles for AI agents, allowing them to operate autonomously without constant human intervention. The concept gained traction in June 2026, building on earlier research such as ReAct (2022) and Reflexion (2023), and represents the latest layer in the progression from prompt engineering to context engineering to harness engineering. This article explores the definition, origin, anatomy, and challenges of building reliable loops.

Agentic AI architecture has evolved by mid-2026 with native reasoning models replacing orchestrated loops, multi-agent swarms, and standardized tool protocols via MCP. This article covers how to design stateless specialist agents, memory graphs, and security patterns.

This article provides a comprehensive guide to building agentic workflows in Python using LangGraph, covering everything from a single model call to a tool-using agent with persistent conversation memory. It explains state, nodes, edges, message history management, model invocation, tool registration and routing, and the ReAct reasoning loop.

This article examines prompt injection and tool misuse threats in agentic AI systems and presents five foundational defense strategies: strict least privilege, open-source guardrails, sandboxed execution, human-in-the-loop checkpoints, and monitoring/auditing.

Learn to install Ollama, download a small language model, and chat completely offline in under 15 minutes. Covers quantization basics and common troubleshooting.

This article explains how scikit-ollama bridges scikit-learn with locally running Ollama models for zero-shot text classification, eliminating the need for cloud APIs. It walks through setting up the environment, loading a movie review sentiment dataset, instantiating a classifier backed by a local Llama 3 model, and using the fit/predict pattern with insights into the underlying mechanism.

A comparison of three open-source LLM evaluation frameworks—RAGAS, DeepEval, and Promptfoo—detailing their purposes, use cases, and the importance of understanding LLM-as-a-judge biases. Includes code walkthroughs for faithfulness checking and CI-gated evaluation.

This article outlines architectural and operational anti-patterns that cause AI agent projects to fail, including premature multi-agent systems, tool sprawl, hardcoded logic, missing memory design, lack of observability, ungoverned write access, context drift, and skipping evaluation. It emphasizes starting simple, building for observability, and adding complexity only when measurable returns justify it.

A comparison of LangChain, LlamaIndex, and raw API calls for LLM applications, covering their strengths, trade-offs, and a decision framework for choosing the right abstraction level.

This article explains how to decide whether agent functionality should be a tool or a subagent, and how to avoid over-engineering. Tools execute code deterministically; subagents execute reasoning in separate contexts. A three-question framework helps make the choice, and the costs of subagents are outlined.

A comprehensive guide on how to maintain tool selection accuracy in AI agents as the tool catalog scales, covering techniques like gating, retrieval-based selection, semantic routing, planning, fallback logic, and benchmarking.

This article explains the distinct roles of context engineering and memory engineering in agentic AI systems. Context engineering focuses on selecting, compressing, and placing information within a single inference call. Memory engineering deals with persisting, retrieving, and maintaining information across calls and sessions. They intersect at the retrieval boundary, where poor management leads to common failures like context pollution or stale data.

This article presents five practical strategies for managing context windows in long-running AI agent applications, along with the key tradeoffs each approach introduces.

Model Context Protocol (MCP) is an open standard by Anthropic that standardizes communication between AI applications and external tools/data sources. This article explains MCP at three levels: why it matters, its architecture and request flow, and production considerations including transport, security, and deployment.

This article explains the seven-layer production AI agent stack, from foundation models to deployment infrastructure. It covers each layer's function, code examples, and technology choices for prototyping, startup, and enterprise scenarios.

This article explains how to distinguish agentic workflows from autonomous agents by focusing on who owns control flow—a human writing code in advance, or a model reasoning at runtime. It covers deterministic workflows, orchestrated workflows, reactive agents (ReAct loop), and autonomous multi-agent systems with runnable code examples. The real axis is predictability versus autonomy, not whether an LLM is involved. Workflows dominate production today, and hybrid architectures are the emerging pattern.

This article explains why large context windows are not the same as agent memory, and how retrieval, compression, and summarization techniques fit together in an agent’s cognitive stack.

Learn how to build a text clustering pipeline using large language model embeddings and HDBSCAN to automatically discover topics in unlabeled text data. Covers embedding generation with sentence-transformers, dimensionality reduction with UMAP, and clustering with HDBSCAN.

This article explains how to build AI agents that can browse and interact with real websites using Playwright, browser-use, and LangGraph. It covers Playwright's advantages over Selenium (30-50% faster, persistent WebSocket, built-in auto-waiting, realistic events), setup steps, dynamic page scraping, multi-step form filling, anti-bot detection handling, session persistence, and Docker deployment. Through code examples, readers will create a working browser agent that navigates sites, fills forms, extracts structured data, and uses an LLM for decision-making.