The following is a learning syllabus generated by AI (and will be modified by me as I learn more).

AI Engineering & Agentic Systems Syllabus

Module 1: The Transformer Architecture & LLM Foundations

• Research Focus: Understand the "Attention" mechanism. This is the heart of every modern LLM.
• Key Readings: Attention Is All You Need (Vaswani et al.). Focus on the Encoder-Decoder structure.
• Concepts: Tokens, Embeddings, Context Windows, and Temperature.
• Learning Goal: Be able to explain how a model predicts the next token and why "context length" is a physical hardware constraint.

Module 2: Retrieval-Augmented Generation (RAG)

• Research Focus: How to solve the "hallucination" problem by giving LLMs a "library" to look at.
• Concepts:
  • Vector Embeddings: Turning text into math.
  • Vector Databases: (Pinecone, Weaviate, or Milvus) for semantic search.
  • Chunking Strategies: How to split a 500-page book so the AI can find the right paragraph.
• Learning Goal: Understand the pipeline: User Query → Embedding → Vector Search → Augmented Prompt → LLM Response.

Module 3: Prompt Engineering & Frameworks

• Research Focus: Moving from basic chatting to programmatic control.
• Tools: LangChain or LlamaIndex.
• Techniques: Chain-of-Thought (CoT), Few-shot prompting, and ReAct (Reason + Act) patterns.
• Learning Goal: Learn how to "chain" multiple AI calls together to solve complex problems that a single prompt can’t handle.

Module 4: Agents and Tool Use (The "Cutting Edge")

• Research Focus: How to give an LLM "hands." (e.g., letting it run a Python script or search the web).
• Concepts: Function Calling, Agentic Loops, and Self-Correction.
• Learning Goal: Understand how an AI determines which tool to use and how it handles errors when a tool returns an unexpected result.

Targeted Projects

To prove you're hireable, you need to show you can handle data and state.

Project 1: The "Compendium Oracle" (RAG System)

Build a local tool that allows you to "talk" to your own tech compendium.

• Tech Stack: Python, LangChain, FAISS (for local vector storage), and OpenAI/Anthropic API.
• The Challenge: Ingest your notes (Markdown/PDFs), store them in a vector DB, and create a CLI tool that answers technical questions based only on your notes.
• Why this works: It demonstrates you understand the most common AI pattern used in corporate environments today.

Project 2: Automated Security Auditor (Agentic Workflow)

Since you have an interest in cybersecurity, build an agent that analyzes code for vulnerabilities.

• Tech Stack: Python, an LLM with Function Calling, and a static analysis tool (like Bandit or Semgrep).
• The Workflow: The agent "reads" a Python file, decides to run a security scan tool, interprets the tool's output, and then writes a "patch" to fix the security flaw.
• Why this works: It shows you can build Agents that interact with the real world (running scripts/writing files), which is the primary direction AI engineering is heading in 2026.

Recommended Learning Resources

• DeepLearning.AI: "Short Courses" on LangChain and AI Agents (very practical).
• Andrejs Karpathy's "Zero to Hero": Specifically the video on building GPT from scratch (the gold standard for theoretical understanding).
• The "Pinecone Learning Center": Excellent for understanding vector math and retrieval.