Maryam Miradi, PhD ✓

[image] After building 400+ Production AI Agents I curated 7 Crucial Skills so you can become a Production AI Agents Engineer. 𝟭. 𝗦𝘆𝘀𝘁𝗲𝗺 𝗗𝗲𝘀𝗶𝗴𝗻 If you have built a backend with multiple services talking to each other, you already think this way. If not, this is where I

[video] 🥇🏆This Is the Most Complete Paper on Agentic RAG I've Read: An Absolute Zero-to-Hero Journey That Explains Everything You Need to Know If you've ever felt overwhelmed by the technical complexity of Retrieval-Augmented Generation (RAG) or thought, “Where do I even begin?”, this

[image] I curated 15 Crucial AI Agent Design Patterns. Most AI engineers don't know half of these exist. After building 400+ AI Agents, this is the AI Agent Architecture Designs I wish I had on day one. + The Link for more Deep Dive. TIER 1: SINGLE-AGENT PATTERNS --- 1. ReAct:

[image] 🛠️🧭 I went down the AI Agent evaluation rabbit hole. If you are building AI Agents for production, you have to evaluate much more than the final answer: tool calls, retrieval quality, reasoning paths, failed steps, latency, cost, safety and regressions. So I went through the

[image] 🛠️🧭 I went down the AI Agent observability rabbit hole. If you are building AI Agents for medical, finance, legal, cybersecurity or other regulated workflows, you also have to deal with: not knowing which tool caused the failure, why the agent retried, which step increased

[image] 🛠️🧭 I went down the AI Agent security rabbit hole. If you are building AI Agents that browse the web, call tools, execute code, access internal systems or act on behalf of users, you also have to deal with: prompt injection, excessive permissions, unsafe tool use, leaked

[image] A very interesting new paper on AI Agent memory. If you build long-horizon agents, this one is worth your time. (bookmark it) Most agent memory systems ask: What should the agent remember? This paper asks a better question: What should it remember, but NOT use right now?

[image] I built an End-to-End AI agent system for Legal documents. 0 Hallucination. Google ADK. 4 parsers incl. Docling. MCP Server. Here is how I build it in 5-step: 1. Business Understanding: Define boundaries before touching code. ☑ Create LegalAgentState with Pydantic

[image] 🛠️🧭 I went down the OCR + PDF extraction rabbit hole. If you are building RAG systems or AI Agents around real documents, you also have to deal with: tables, formulas, reading order, scanned pages, multi-column layouts, charts, headers and footnotes. So I went through the

[image] What if AI actually reached the 𝗽𝗮𝘁𝗶𝗲𝗻𝘁 𝗯𝗲𝗱𝘀𝗶𝗱𝗲? Here is how to build a $𝟭.𝟱𝗠 Medical AI Agent pipeline. ⬇️ Most medical AI dies in deployment. Why? + Clinicians don't trust black-box AI. + Data scientists waste 80% of their time cleaning messy data. While

[image] 🛠️🧭 How to Build AI Agent Harnesses That Improve Across 6 Dimensions. 𝗠𝘆 𝟲-𝘀𝘁𝗲𝗽 𝗿𝗼𝗮𝗱𝗺𝗮𝗽 𝘁𝗼 𝗔𝗱𝗮𝗽𝘁𝗶𝘃𝗲 𝗔𝗜 𝗔𝗴𝗲𝗻𝘁 𝗵𝗮𝗿𝗻𝗲𝘀𝘀𝗲𝘀. Most AI agent harnesses have a static problem: We configure the prompt, tools, memory, workflow, and output handling

[image] 🛠️🧭How to Build AI Agents with Claude Code → Even If You're Starting from Zero. 𝗧𝗵𝗶𝘀 𝗶𝘀 𝟵 𝗳𝗲𝗮𝘁𝘂𝗿𝗲 𝗿𝗼𝗮𝗱𝗺𝗮𝗽 𝗳𝗿𝗼𝗺 𝗖𝗟𝗔𝗨𝗗𝗘.𝗺𝗱 𝘁𝗼 𝗠𝗖𝗣 𝗦𝗲𝗿𝘃𝗲𝗿. 》𝗙𝗲𝗮𝘁𝘂𝗿𝗲 𝟭: CLAUDE. md Write this before any code. Not after setup. Before. ✸ Most

[image] 🛠️🧭How to Build AI Agents with Claude Code → Even If You're Starting from Zero. 𝗧𝗵𝗶𝘀 𝗶𝘀 𝟵 𝗳𝗲𝗮𝘁𝘂𝗿𝗲 𝗿𝗼𝗮𝗱𝗺𝗮𝗽 𝗳𝗿𝗼𝗺 CLAUDE. md 𝘁𝗼 𝗠𝗖𝗣 𝗦𝗲𝗿𝘃𝗲𝗿. 》𝗙𝗲𝗮𝘁𝘂𝗿𝗲 𝟭: CLAUDE. md Write this before any code. Not after setup. Before. ✸ Most critical

[image] 🛠️🧭 How to do Harness Engineering for AI Agents. 𝗧𝗵𝗶𝘀 𝗶𝘀 𝗮 𝟳-𝗦𝘁𝗲𝗽 𝗿𝗼𝗮𝗱𝗺𝗮𝗽 𝗳𝗿𝗼𝗺 𝗰𝗼𝗱𝗶𝗻𝗴 𝘁𝗮𝘀𝗸𝘀 𝘁𝗼 𝘀𝘆𝘀𝘁𝗲𝗺 𝘀𝘁𝗲𝗲𝗿𝗶𝗻𝗴. In the era of Claude Code and Codex, the bottleneck is moving from writing code, reviewing every diff to

[image] 🛠️🧭 How to Use Graph Engineering to Build Graph-Powered Road AI Agents. 𝗠𝘆 𝟳-𝘀𝘁𝗲𝗽 𝗿𝗼𝗮𝗱𝗺𝗮𝗽 𝗳𝗿𝗼𝗺 𝗮 𝗿𝗮𝘄 𝗿𝗼𝗮𝗱 𝗻𝗲𝘁𝘄𝗼𝗿𝗸 𝘁𝗼 𝗴𝗿𝗮𝗽𝗵-𝗽𝗼𝘄𝗲𝗿𝗲𝗱 𝗽𝗿𝗼𝗱𝘂𝗰𝘁𝗶𝗼𝗻 𝗔𝗜 𝗮𝗴𝗲𝗻𝘁𝘀. Most agents are good at reading properties. Condition:

[image] Don't Build AI Agents without Learning These 20 Fundamentals. Level 1 to Level 5: After building 400+ agentic systems, these are the 20 concepts I think matter most: Level 1: CORE REASONING --- 1. Agent Loops 2. Deterministic vs. Probabilistic Logic 3. Task Decomposition 4.

[image] 🛠️🧭 Jev, TypeSafe AI, Explained. 𝗠𝘆 𝟲-𝘀𝘁𝗲𝗽 𝗝𝗲𝘃 (𝗥𝗟𝗛𝗙 𝘁𝗼 𝗥𝗟𝗖𝗗). Diogo Almeida helped build the RLHF/InstructGPT work behind the ChatGPT era. Now his company TypeSafe has released Jev. And the idea is very different: Instead of building another model

[image] I curated 15 Crucial AI Agent Design Patterns. Most AI engineers don't know half of these exist. After building 400+ AI Agents, this is the AI Agent Architecture Designs I wish I had on day one. + The Link for more Deep Dive. TIER 1: SINGLE-AGENT PATTERNS --- 1. ReAct:

[image] 🛠️🧭 I went down the AI Agent memory rabbit hole. If you are building AI Agents that need to remember users, past actions, preferences or changing facts across sessions, you also have to deal with: short-term state, long-term memory, memory retrieval, stale facts, conflicting

[image] 🛠️🧭 How to Review AI-Written Code Like a Senior AI Engineer. 𝗠𝘆 𝟲-𝘀𝘁𝗲𝗽 𝗽𝗮𝘁𝗵 𝘁𝗼 𝗽𝗿𝗼𝗱𝘂𝗰𝘁𝗶𝗼𝗻-𝗿𝗲𝗮𝗱𝘆 𝗰𝗼𝗱𝗲. AI coding agents can now generate more code than one engineer can realistically read line by line. So we need to review by risk, not by

[image] 🛠️🧭 How to Improve Production AI Agents With Continual Learning. 𝗠𝘆 𝟱-𝘀𝘁𝗲𝗽 𝗿𝗼𝗮𝗱𝗺𝗮𝗽 𝗳𝗿𝗼𝗺 𝘀𝘁𝗮𝘁𝗶𝗰 𝗔𝗜 𝗔𝗴𝗲𝗻𝘁𝘀 𝘁𝗼 𝗖𝗼𝗻𝘁𝗿𝗼𝗹𝗹𝗲𝗱 𝗖𝗼𝗻𝘁𝗶𝗻𝘂𝗮𝗹 𝗟𝗲𝗮𝗿𝗻𝗶𝗻𝗴. Most production AI Agents have a hidden weakness: They can fail, get

[image] 🧭 The path from AI coding to production AI agents is full of gaps. → Atlassian is tackling 6 serious ones with connected context. Here’s how. 》The Gap that Needed Closing I've built 400+ production AI agents. AI coding is making them faster to build. And I can tell you:

[image] 🛠️🧭 New tool ≠ New architectures. Tools come and go. Agent Engineering stays. 𝗠𝘆 𝟭𝟮-𝗹𝗮𝘆𝗲𝗿 𝗲𝗻𝗴𝗶𝗻𝗲𝗲𝗿𝗶𝗻𝗴 𝗰𝗼𝗻𝘁𝗲𝘅𝘁 𝘁𝗼 𝗰𝗼𝗻𝘁𝗶𝗻𝘂𝗮𝗹. AI coding has made it incredibly easy to build agents. But after building 400+ production AI agents, I kept

[image] 🛠️🧭 Workflow vs Agent vs Multi-Agent vs Agentic System. 𝗠𝘆 𝟰-𝗽𝗮𝗿𝘁: 𝗙𝗹𝗼𝘄 → 𝗗𝗲𝗰𝗶𝘀𝗶𝗼𝗻 → 𝗖𝗼𝗹𝗹𝗮𝗯𝗼𝗿𝗮𝘁𝗶𝗼𝗻 → 𝗦𝘆𝘀𝘁𝗲𝗺. Imagine your production checkout suddenly returns: 500 Internal Server Error You could solve this with a workflow. Or an agent.

[image] 🛠️🧭 How to Build Self-Evolving AI Agents Without Letting Them Grade Themselves. 𝗠𝘆 𝟳-𝘀𝘁𝗲𝗽 𝗿𝗼𝗮𝗱𝗺𝗮𝗽 𝘁𝗼 𝗲𝘃𝗶𝗱𝗲𝗻𝗰𝗲-𝗴𝗮𝘁𝗲𝗱 𝗔𝗜 𝗔𝗴𝗲𝗻𝘁𝘀 Most “self-improving” agents have a dangerous loop: The agent proposes a change, tests it and decides it worked.

[image] 🛠️🧭 How to Build OmniScientist: An Omni-Modal Multi-Agent Research Automation. 𝗠𝘆 𝟳-𝗦𝘁𝗲𝗽 𝗿𝗼𝗮𝗱𝗺𝗮𝗽 𝗳𝗿𝗼𝗺 𝗼𝗺𝗻𝗶-𝗺𝗼𝗱𝗮𝗹 𝘀𝗰𝗶𝗲𝗻𝘁𝗶𝗳𝗶𝗰 𝗱𝗮𝘁𝗮 𝘁𝗼 𝗮𝘂𝘁𝗼𝗻𝗼𝗺𝗼𝘂𝘀 𝘀𝗰𝗶𝗲𝗻𝘁𝗶𝗳𝗶𝗰 𝗱𝗶𝘀𝗰𝗼𝘃𝗲𝗿𝘆. Current AI Scientist agents can generate

[image] A VLM Can Read One Medical Screen. A Medical AI Agent Must Finish 24 Screens in a Row. 𝗧𝗵𝗶𝘀 𝗶𝘀 𝗺𝘆 𝟳-𝗦𝘁𝗲𝗽 𝗿𝗼𝗮𝗱𝗺𝗮𝗽 𝗳𝗿𝗼𝗺 𝗦𝗰𝗿𝗲𝗲𝗻-𝗮𝘀-𝗦𝘁𝗮𝘁𝗲 𝘁𝗼 𝗮 𝗽𝗿𝗼𝗱𝘂𝗰𝘁𝗶𝗼𝗻 𝗠𝗲𝗱𝗶𝗰𝗮𝗹 𝗔𝗜 𝗔𝗴𝗲𝗻𝘁. Real medical agent must move through many

[image] 🛠️🧭 The Art of Loop Engineering: How to Build AI Agents That Improve Over Time. 𝗧𝗵𝗶𝘀 𝗶𝘀 𝗺𝘆 𝟰-𝗟𝗼𝗼𝗽 𝗳𝗿𝗮𝗺𝗲𝘄𝗼𝗿𝗸 𝗳𝗿𝗼𝗺 𝗼𝗻𝗲 𝗮𝗴𝗲𝗻𝘁 𝗮𝘁𝘁𝗲𝗺𝗽𝘁 𝘁𝗼 𝗮 𝘀𝘆𝘀𝘁𝗲𝗺 𝘁𝗵𝗮𝘁 𝗴𝗲𝘁𝘀 𝗯𝗲𝘁𝘁𝗲𝗿 𝗼𝘃𝗲𝗿 𝘁𝗶𝗺𝗲. Loops are not new. Software has

[image] 🛠️🧭 My 9 Production AI Agent Layers That Claude Code Doesn’t Replace. 𝗙𝗿𝗼𝗺 𝗮 𝗖𝗹𝗮𝘂𝗱𝗲 𝗣𝗿𝗼𝘁𝗼𝘁𝘆𝗽𝗲 𝘁𝗼 𝗣𝗿𝗼𝗱𝘂𝗰𝘁𝗶𝗼𝗻 𝗔𝗜 𝗔𝗴𝗲𝗻𝘁. Claude Code can get you to a working prototype incredibly fast. But production is where the harder engineering starts.

[image] 🛠️🧭 You Can Learn Production AI Agents in 20 Minutes. 𝗠𝘆 𝟭𝟮-𝗹𝗮𝘆𝗲𝗿 𝗲𝗻𝗴𝗶𝗻𝗲𝗲𝗿𝗶𝗻𝗴 𝘀𝘁𝗮𝗰𝗸 𝗳𝗿𝗼𝗺 𝗰𝗼𝗻𝘁𝗲𝘅𝘁 𝘁𝗼 𝗰𝗼𝗻𝘁𝗶𝗻𝘂𝗮𝗹 𝗹𝗲𝗮𝗿𝗻𝗶𝗻𝗴. AI coding has made it incredibly easy to build agents. But after building 400+ production AI agents, I

[image] 🛠️🧭 AI Agents Are Not Enough. Real AI Engineers Build Software Factories. 𝗠𝘆 𝟲-𝗟𝗮𝘆𝗲𝗿 𝗥𝗼𝗮𝗱𝗺𝗮𝗽 𝘁𝗼 𝗮𝗻 𝗔𝘂𝘁𝗼𝗻𝗼𝗺𝗼𝘂𝘀 𝗦𝗼𝗳𝘁𝘄𝗮𝗿𝗲 𝗙𝗮𝗰𝘁𝗼𝗿𝘆. A swarm of coding agents is still not a software factory. Generating code is only one part of software

[image] 🛠️🧭 Skills vs MCP vs RAG vs Memory What AI Agents Need to Know. 𝗠𝘆 𝟰-𝗽𝗮𝗿𝘁: 𝗣𝗿𝗼𝗰𝗲𝗱𝘂𝗿𝗲 → 𝗔𝗰𝗰𝗲𝘀𝘀 → 𝗞𝗻𝗼𝘄𝗹𝗲𝗱𝗴𝗲 → 𝗘𝘅𝗽𝗲𝗿𝗶𝗲𝗻𝗰𝗲. Imagine your production checkout suddenly returns: 500 Internal Server Error A common approach is to dump

[image] 🛠️🧭 Stop Building AI Agents Like Coding Agents. 𝗠𝘆 𝟲-𝗹𝗮𝘆𝗲𝗿 𝘁𝗼 𝗴𝗿𝗼𝘂𝗻𝗱𝗲𝗱 𝗞𝗻𝗼𝘄𝗹𝗲𝗱𝗴𝗲 𝗪𝗼𝗿𝗸. Most AI Agent architecture today is heavily influenced by coding agents. But code is a very special kind of knowledge. It has filenames, identifiers,