AI/ML Engineer Path
From Python fundamentals through classical ML, deep learning, RAG, and agentic AI to the specific AI job families hiring through 2030.
Click a node to expand it — summary, free resources, and a checkbox to mark it complete. Progress is saved on this device only.
AI/ML Engineer Path
17 nodes · 0 marked complete
The AI timeline, 2020-2026
From GPT-3 (2020) through ChatGPT (Nov 2022), Claude/GPT-4 (2023), and the 2025-2026 agent era (Claude Code, Copilot Agent), tooling adoption is near-universal. But measured productivity gains have been smaller and later than the hype — a controlled study found experienced developers 19% slower with early-2025 AI tools on complex tasks, though a 2026 follow-up shows that gap closing as tools matured.
Python fundamentals (prerequisite)
Every roadmap on this page assumes Python Stage 1-2 fluency (syntax, standard library, basic tooling) — the shared entry ticket for classical ML, deep learning, and the application-first GenAI track alike. If you are starting from a typed language like TypeScript, this compresses to about 3 weeks.
Math minimum for ML
Linear algebra (vectors, matrices, dot products), basic calculus (derivatives, gradients), and probability/statistics (distributions, Bayes, mean/variance) — the minimum viable math to understand why the algorithms in the next tier work, not a full math degree.
Generative AI & API fluency
The fastest path to something useful: prompt engineering (zero/few-shot, chain-of-thought, structured output), and OpenAI/Anthropic SDK fluency (chat completions, streaming, tool calling, cost control). This "application-first" track can start immediately, without classical ML or deep learning first, and is the standard 2026 route for working developers.
Data engineering foundations
Data pipelines feeding both analytics and AI are the unglamorous bottleneck — and Big Data Specialist is the #1 fastest-growing role on the WEF 2025 list. SQL, Spark/dbt, streaming, and vector-store basics underpin every other track on this page.
Classical machine learning
Supervised vs. unsupervised learning, train/validation/test splits, overfitting, and the core algorithm family (regression, decision trees, gradient boosting, k-means, PCA) via scikit-learn end-to-end pipelines. Roughly 3-4 months to a working Kaggle-competition level.
RAG systems
Retrieval-Augmented Generation grounds a frozen, hallucination-prone LLM in your own documents: chunking, embeddings, vector stores (pgvector/Pinecone/Qdrant), hybrid search + reranking, and RAGAS-style evaluation. About 1-2 months, and the backbone of most "chat with our data" products.
Deep learning
Neural net fundamentals (backprop, gradient descent variants), PyTorch as the industry-default framework, and the architecture progression MLP → CNN → Transformer. Practice on MNIST/CIFAR, then fine-tune a pretrained model. About 3-4 months.
Agentic AI
An agent is an LLM in a loop with tools, memory, and a goal (the ReAct pattern). Start with raw tool-calling (no framework) to understand the loop, then LangGraph or the OpenAI Agents SDK, plus MCP for connecting agents to external tools/data. About 2-3 months; building and supervising these systems is the premium specialization of the late 2020s.
NLP specialization
Tokenization, embeddings (word2vec → contextual → sentence embeddings), and transformer-based NLP (BERT vs. GPT families) via Hugging Face `transformers`. About 2 months — worth it specifically if you are targeting NLP-specialist roles, optional otherwise.
LLM internals
Popping the hood: next-token prediction, tokenizers (BPE), the pretrain → SFT → RLHF/DPO pipeline, why models hallucinate, the open-weights ecosystem (Llama/Mistral/Qwen via Ollama), and when (mostly NOT) to fine-tune with LoRA/QLoRA. About 2-3 months.
Role: AI/LLM Application Engineer
Builds products on top of foundation models — chat, copilots, extraction, search. Job postings exploded 2024-2026 and "AI engineer" is now a standard title on LinkedIn/Naukri. Closest transition for today's application developers, and the role most likely to simply become the default definition of "software engineer" by 2030.
Role: Agentic Systems Engineer
Designs multi-step autonomous workflows — agents, tools, memory, and guardrails. 2025-26 job postings list "agent orchestration, MCP integration, eval design" as top differentiators, and every major vendor shipped an agent SDK in 2025. High growth; the premium specialization of the late 2020s.
Role: MLOps / AI Platform Engineer
Productionizes and operates models: serving, monitoring, pipelines, GPU cost optimization. Every AI feature a company ships needs someone to run it — a reported +20-40% pay premium over generalist AI roles in the Indian market, and a role that consolidates durably with platform engineering.
Role: AI Evaluation / Safety Engineer
Builds test harnesses, red-teams models, and enforces guardrails and compliance. Eval design is reported as one of the strongest signals of real LLM experience in 2026 hiring guides, and regulation (e.g. the EU AI Act) is turning this from nice-to-have into a formal, often mandatory function in regulated industries.
Role: ML Engineer / Data Scientist
The pre-ChatGPT AI job: training and tuning predictive models on tabular and domain data for fraud, demand, or credit-risk problems. Less flashy than LLM work, but WEF ranks AI/ML Specialists and Big Data Specialists among the very top growth roles through 2030 — steady demand in fintech, health, and logistics.
Role: AI Product Engineer
Ships user-facing AI features end-to-end — frontend, backend, LLM APIs, and UX for non-determinism, all in one person. Startups increasingly hire this over separate frontend/backend/ML trios; the safest, highest-leverage evolution path for a current full-stack developer.
Node list (accessible fallback)
| Tier | Node | Prerequisites |
|---|---|---|
| 0 | The AI timeline, 2020-2026 | — |
| 0 | Python fundamentals (prerequisite) | — |
| 1 | Math minimum for ML | Python fundamentals (prerequisite) |
| 1 | Generative AI & API fluency | Python fundamentals (prerequisite) |
| 2 | Data engineering foundations | Python fundamentals (prerequisite) |
| 2 | Classical machine learning | Math minimum for ML |
| 2 | RAG systems | Generative AI & API fluency |
| 3 | Deep learning | Classical machine learning |
| 3 | Agentic AI | RAG systems, Generative AI & API fluency |
| 4 | NLP specialization | Deep learning |
| 4 | LLM internals | Deep learning, Generative AI & API fluency |
| 5 | Role: AI/LLM Application Engineer | RAG systems, Generative AI & API fluency |
| 5 | Role: Agentic Systems Engineer | Agentic AI |
| 5 | Role: MLOps / AI Platform Engineer | Classical machine learning |
| 5 | Role: AI Evaluation / Safety Engineer | LLM internals |
| 5 | Role: ML Engineer / Data Scientist | Classical machine learning |
| 6 | Role: AI Product Engineer | Role: AI/LLM Application Engineer |