Self-Study AI Skill Development Roadmap

[!INFO]
This is my early AI skill development roadmap that I followed for myself. Hope, it helps for you as well in case of you are looking for courses and resources. I also added a quick and smooth 4‑week sprint plan for a simple start. My recommendation is loving to read, if you don't like it. Because this field is living with reading! You need to also develop muscles for interpreting arxiv.org papers!

Quick wins (free, short, high‑signal)

• Generative AI with LLMs (DeepLearning.AI × AWS) — crisp foundations + deployment mindset.
• Link: https://www.coursera.org/learn/generative-ai-with-llms/
• Generative AI Foundations on AWS — hands-on pretrain/fine‑tune/deploy on Bedrock; pairs well with AWS workflows.
• Link: https://aws.amazon.com/blogs/machine-learning/new-technical-deep-dive-course-generative-ai-foundations-on-aws/

Core learning paths

A. Practical builder track (LLMs → apps → ops)

1. Intro to ML in Production (MLOps lens) — adopt the “ship it safely” mindset early.
• Link: https://www.coursera.org/learn/introduction-to-machine-learning-in-production/
2. Generative AI with LLMs — modern LLM patterns (prompting, RAG, evals, costs).
• Link: https://www.coursera.org/learn/generative-ai-with-llms/
3. Agent frameworks
• LangGraph agents — controllable agent graphs, memory, human‑in‑the‑loop.
• Docs: https://langchain-ai.github.io/langgraph/reference/agents/
• LlamaIndex agentic workflows — document/knowledge agents & multi‑agent workflows.
• Docs: https://docs.llamaindex.ai/ and https://docs.llamaindex.ai/en/latest/module_guides/workflow/
• Transformers Agents 2.0 → smolagents — transparent tool‑using agents; see the HF blog and current API notes.
• Blog: https://huggingface.co/blog/agents
• Docs (status): https://huggingface.co/docs/transformers/agents
4. RL for LLMs (when you go beyond SFT) — TRL / reinforcement libraries and examples.
• Starter: https://github.com/huggingface/trl

B. Theory depth (for red‑teamers who also build)

• CMU 11‑785 Intro to Deep Learning — current lectures & assignments.
• Site: https://deeplearning.cs.cmu.edu/
• Stanford CS224N (NLP/LLMs) — foundations + updated deployment/efficiency material.
• Site: http://web.stanford.edu/class/cs224n/
• MIT 6.S191 — short, modern overviews of deep learning and LMs.
• Site: http://introtodeeplearning.com/

[!TIP]
Alternate a builder week (LangGraph/LlamaIndex) with a theory week (CS224N notes).

Security canon (bookmark these)

[!IMPORTANT]
Keep these references close; they’re the “north star” while you iterate on attacks, defenses, and evaluations.

• OWASP GenAI Security Project & LLM Top 10 (v1.1 / 2025) — current risk taxonomy and resources.
• Hub: https://owasp.org/www-project-top-10-for-large-language-model-applications/
• OWASP ML Security Top 10 — classic ML risks beyond LLMs (poisoning/evasion/etc.).
• Link: https://owasp.org/www-project-machine-learning-security-top-10/
• MITRE ATLAS — adversary TTPs mapped across the AI lifecycle.
• Link: https://atlas.mitre.org/
• NIST AI RMF — Generative AI Profile (2024) — governance & control families that map cleanly to GenAI systems.
• PDF: https://doi.org/10.6028/NIST.AI.600-1

Prompt‑security & modern injection variants

• NCC Group – Exploring Prompt Injection — foundational read; pair with newer role‑targeted variants.
• Link: https://research.nccgroup.com/2022/12/05/exploring-prompt-injection-attacks/
• Context for stakeholders — indirect injection explainers and case studies (add to briefings).

Safe playgrounds to practice breaking

• Lakera Gandalf (progressive defenses, input/output guardrails).
• Play: https://www.lakera.ai/lakera-gandalf
• Explainer: https://www.lakera.ai/insights/who-is-gandalf
• DoubleSpeak — jailbreak & detection exercises.
• Play: https://doublespeak.chat
• Handbook: https://doublespeak.chat/#/handbook

[!TIP]
Practice loop to log: attack surface → payload → effect → mitigation(s) tried → retest result.

Agents & tools

• Focus your effort on: tool use, retrieval policies, output‑parsing, eval harnesses, and agent memory hardening.
• LangGraph agents & templates: https://langchain-ai.github.io/langgraph/reference/agents/
• LlamaIndex Workflows & Agent Workflows: https://docs.llamaindex.ai/en/latest/module_guides/workflow/ and https://docs.llamaindex.ai/en/latest/understanding/agent/multi_agents/
• HF Agents 2.0 (now spun out to smolagents): https://huggingface.co/blog/agents and https://huggingface.co/docs/transformers/agents

Cloud context (AWS‑centric)

[!INFO]
The links below bias to AWS so you can quickly map learning to hands‑on deployments.

• AWS GenAI training hub — rolling updates & cert‑level paths.
• Start here: https://aws.amazon.com/training/learn-about/generative-ai/
• Bedrock model landscape — keep track of the latest foundation/reasoning models and inference profiles via AWS docs.

Suggested weekly progression (4 weeks, security‑first)

[!NOTE]
Treat this as a flexible cadence—swap weeks or extend topics based on your background and current projects.

• DeepLearning.AI Generative AI with LLMs modules 1–2 + AWS Generative AI Foundations labs.
• Read OWASP GenAI Top 10 summaries; map each risk to one Bedrock use case you know.

Week 1 — Foundations

• Build one LangGraph agent with a web‑search tool; add LlamaIndex for retrieval.
• Draft a mini threat model using MITRE ATLAS tactics for tool use, retrieval, and output handling.

Week 2 — Agents + Threat modeling

• Run Gandalf + DoubleSpeak; reproduce classic NCC prompt injection and try role‑targeted variants.
• Capture mitigations and evaluation signals (e.g., refusal‑rate deltas, tool‑call diffs).

Week 3 — Red‑teaming drills

• Align controls with NIST AI RMF + GenAI Profile; write a short “policy‑to‑prompting” map (what guardrail implies what test).

Week 4 — Governance & guardrails

Recommended YouTube Channels

(Thanks, Garrett !)

• https://www.youtube.com/@matthew_berman
• https://www.youtube.com/@mreflow
• https://www.youtube.com/@YannicKilcher
• https://www.youtube.com/@HuggingFace
• https://www.youtube.com/@RobertMilesAI
• https://www.youtube.com/@HeatonResearch
• https://www.youtube.com/@NicholasRenotte
• https://www.youtube.com/@reidhoffman
• https://www.youtube.com/@testingai/videos
• https://www.youtube.com/@engineerprompt
• https://www.youtube.com/@MachineLearningStreetTalk
• https://www.youtube.com/@Deeplearningai/videos
• https://www.youtube.com/@DrAlanDThompson

-EOF