I made a small test set to check it and got my first real scores. Most cases passed, a couple failed, and those are what I’m fixing next. Good to have actual numbers to beat now instead of guessing.

The environment is set up and the first version of my project is running. The core engine is working and the first AI agent is already reading live data. Next up: connecting the rest of the AI and testing it properly. More soon.

Just wrapped the “From Chatbot to Decision Engine: Build Agentic AI Live” masterclass with Mohammad Arshad at Decoding Data Science. We built a Gemini agent for a busy airport. It reads policy documents with RAG, pulls hourly queue data through fixed SQL tools, and answers with evidence and citations. The demo question was “Is Terminal 1 meeting the departure immigration standard on Friday evening?” The answer was no: 36.1 min actual vs. a 15 min standard. My key takeaways: • Prompt → Context → Harness → Loop. Engineer and test all four. • A prompt alone is not a security boundary. Read-only tools with fixed queries are. • The agent routes on the tool description, so write it well. • Plan for safe failure, like a fallback model. I also took 1st place 🥇 in the Agentic AI Kahoot Quiz (6/6, 5,487 pts), with Mohammad Umar 2nd 🥈 and MA 3rd 🥉. Thanks Decoding Data Science and Mohammad Arshad! #AgenticAI #DecodingDataScience #AgenticAIFestival

Finished the Builder Codex and took the pledge. My main takeaway: stop collecting courses and start building. I'm putting that into practice in the AI Application Building Challenge with BlackRoute, an agentic AI app that spots logistics disruptions early and helps plan the recovery. I'll share progress here as I build.