AI Learning 2026
A 12-week AI learning sprint (Oct 2026 – Jan 2027): agentic AI, MCP, and AI for systems engineering. Every milestone ships code here, and every course produces public content.
Follow along: vibewithamit.com · TikTok @vibewithamitai · YouTube @vibewithamitai · Instagram @vibewithamitai, with videos, posts and write-ups from each week.
Progress
| Week | Course | Milestone | Status |
|---|
| 1 | Anthropic Academy | Claude Code in Action certificate | ✅ Done |
| 2 | Anthropic Academy | API tool-use script | ✅ Done |
| 3 | Anthropic Academy | Custom MCP server | ✅ Done |
| 4–8 | Agentic AI (Andrew Ng) | Reflection, tool use, evals, multi-agent | ⬜ |
| 8 | PPI AI for Systems Engineering | Workshop attended | ⬜ |
| 9–11 | Hugging Face AI Agents Course | Agent rebuilt in LangGraph, certificate | ⬜ |
| 12 | Capstone | Requirements traceability agent demo | ⬜ |
Structure
01-anthropic-academy/ Claude Code, Claude API, MCP projects
02-agentic-ai/ Agent design patterns and evals (DeepLearning.AI)
03-huggingface-agents/ smolagents and LangGraph work
04-capstone/ Requirements traceability agent (synthetic data only)
content/ Links to videos, posts and blog articles
website/ Vibe with Amit site (vibewithamit.com), deployed to Netlify
Highlights
- MCP server (Week 3): the requirements checker's tools packaged as an MCP server and connected to Claude Code, which called them 9 times in one review, then pointed out gaps in the rules themselves. Code
- Requirements checker (Week 2): Claude used two Python tools to review 8 synthetic system requirements, flagged 6 (vague wording, weak modal verbs, a TBD, a compound requirement) and rewrote each as testable, marking invented values as [assumed]. Code · sample output
- Prompt v1 vs v2 (Week 2): four prompt engineering techniques took the eval from 5/6 to 6/6 across two runs, with answers 39% shorter and quality up from 8.7 to 8.8. Code
- Eval caught the AI grader (Week 2): a 9/10 answer broke the 120-word limit; a one-line code check caught it. Code
Certificates
Note
All examples use synthetic or public data only.
