Agentic AI, LLM engineering, and the models behind modern automation — multi-agent systems, LLM evaluation and comparisons, RAG, fine-tuning, AI infrastructure, security, and production AI engineering.
From the blog
Build an AI agent that reads documents, extracts named entities and their relationships, constructs a knowledge graph stored in Neo4j, and provides a natural language query interface over the graph.
Build an AI agent that performs structured legal reasoning — searching precedents, constructing multi-step arguments with evidence chains, generating counter-arguments, and producing balanced legal analysis.
Learn how to build AI agents with metacognitive capabilities — uncertainty estimation, confidence calibration, knowledge boundary detection, and know-when-to-ask patterns that make agents more reliable and honest.
Understand the MRKL (Modular Reasoning, Knowledge, and Language) architecture that combines LLMs with specialized expert modules, intelligent routing, and structured knowledge retrieval for building powerful AI systems.
Explore multi-agent reinforcement learning (MARL) concepts including reward shaping, cooperative versus competitive strategies, and policy gradient methods for agent teams with practical Python implementations.
Construct a multi-agent research system with four specialized agents — Scientist, Librarian, Analyst, and Writer — that collaborate on source discovery, analysis, and paper generation with complete Python code.
Learn how to build agent systems that route different reasoning tasks to different language models — using fast, cheap models for classification and routing, and powerful models for generation and complex reasoning.
Master the technical details of resolution management, coordinate scaling, multi-monitor awareness, and viewport configuration for reliable Claude Computer Use agents across different display setups.
Build GPT Vision agents that handle complex multi-step web workflows spanning multiple pages. Learn task decomposition, state tracking, page transition handling, and verification at each step.
