Subscribe
Sign in
Home
Notes
Chat
LLM Engineer's Handbook
Agentic AI Engineering Course
Roadmaps
Perks
Contact Me
Archive
About
Foundations
Latest
Top
Discussions
Agentic Harness Engineering
Building systems that transform the LLM into the new operating system
Mar 31
•
Paul Iusztin
54
12
9
From 12 Agents to 1
The mental model that prevents you from overengineering your next AI system.
Mar 26
•
Paul Iusztin
and
Louis-François Bouchard
42
2
9
The AI Evals Roadmap I Wish I Had
From vibe checking to trusted agents in production
Mar 24
•
Paul Iusztin
63
6
9
Why RAG Has Exactly 6 Failure Modes. No More, No Less.
A complete guide for evaluating your retrieval-augmented generation systems.
Mar 17
•
Paul Iusztin
32
6
3
Our LLM Judge Passed Everything. It Was Wrong.
Align your evaluator with human judgment, or don't trust it at all.
Mar 10
•
Paul Iusztin
20
7
4
How to Design Evaluators That Catch What Actually Breaks
The practical guide to code-based checks, LLM judges, and rubrics for real-world AI apps
Mar 3
•
Paolo Perrone
22
6
5
Generate Synthetic Datasets for AI Evals
5 strategies from cold start to 450 diverse inputs in minutes
Feb 24
•
Paul Iusztin
26
7
5
No Evals Dataset? Here's How to Build One from Scratch
Build evaluators to signal problems that users actually care about. Step-by-step guide.
Feb 17
•
Paul Iusztin
29
1
4
Integrating AI Evals Into Your AI App
The holistic guide: From optimization to production monitoring
Feb 10
•
Paul Iusztin
41
7
8
How to Design Python AI Projects That Don't Fall Apart
A framework-agnostic approach to modular agents, workflows, and LLM apps using the pragmatic clean architecture design.
Jan 13
•
Paul Iusztin
61
4
12
The AI Agents Roadmap Nobody Is Teaching You
9 lessons: from workflows to planning and multimodal AI agents. From scratch. Framework free.
Dec 16, 2025
•
Paul Iusztin
134
6
17
Stop Converting Documents to Text. You're Doing It Wrong.
How to work with multimodal agents: images, PDFs, audio, and... text.
Dec 9, 2025
•
Paul Iusztin
48
11
7
This site requires JavaScript to run correctly. Please
turn on JavaScript
or unblock scripts
