Back to portfolio

my career transition · v2026

UI Engineer → AI Engineer
a map I built for myself, not generated for everyone

I am a frontend engineer with 12+ years in UI/Product, and this is the route I am following to add AI engineering to my toolkit. My frontend background is not a weakness here — it is the multiplier. I am not starting from zero; I am bolting an AI layer on top of skills I already have: React, state management, API contracts, evals, and shipping UI at scale. The phases below are what I actually study, the projects I actually build, and the resources I keep open on my desk.

free paid bookP0…P6 — stations I am tracking

Resources go stale fast —This is my personal filter. I hold on to principles — RAG, evals, agents, orchestration — and let specific API versions change underneath me. A 2024 course without agents and MCP is already dated; the anchor books are about fundamentals, not news.

P0 · MONTH 1

Foundation: Python + how LLMs think

I did not come here to learn programming from scratch. I came here to port the way I already think — async, state, API contracts — into Python. My first goal is to rewrite one of my own JS utilities in FastAPI and understand how an LLM behaves as a dependency, not a magic box.

AI Engineering Roadmap 2026

The 13-step meta map I keep open on my second monitor. It helps me see where each skill fits.

Andrej Karpathy (YouTube)

I use his mental-model videos instead of a textbook. Understanding beats memorizing.

The async event loop is familiar; this is my Python backend surface.

Anthropic + OpenAI docs

Primary source material. I read API refs before I read tutorials.

Chip Huyen — AI Engineering

The anchor book I treat as my reference. Author is ex-Netflix; that resonated with my own background.

aie-book (repo)

Open companion materials I skim whenever I need a second explanation.

buildPort one of my own JS utilities to FastAPI and make my first tool call. Then write about it so the learning sticks.

P1 · MONTHS 1–2

LLM APIs + prompt engineering

I want to stop 'trying prompts' and start treating LLMs as a production dependency: function calling, structured output with Pydantic, streaming, retries, and token accounting. This is where I get my API discipline right.

DeepLearning.AI — Short Courses

Short, dense modules on prompting, functions, and structured output. I take them in small sprints.

Anthropic Prompt Engineering

The canonical guide. I keep it open when I design any prompt.

Ed Donner — LLM Engineering

14 projects I use as reference implementations and interview-portfolio fodder.

P2 · MONTH 2

RAG — the backbone of most job specs

Almost every AI-engineering job posting I see mentions RAG. I am building it from scratch: chunking, embeddings, vector stores, hybrid search, and reranking. This is where frontend-to-backend integration becomes real.

Hugging Face — LLM Course

My open-source reference. I come here for the model ecosystem, not just the theory.

DL.AI — Advanced RAG / Vector DB

I use these to map retrieval, reranking, and RAG evaluation to a real codebase.

Vector search inside Postgres. I am already comfortable with relational data, so this is my home base.

I keep this as the alternative when I need a dedicated vector engine.

Hands-On LLMs (Alammar)

Visual and hands-on. The embeddings and RAG sections match the way I learn.

build · project 1A streaming chat with a RAG backend. The React frontend is where I move fast; the retrieval layer is where I prove the skill.

P3 · MONTH 3

Evals + MLOps — my differentiator

I noticed that most junior AI resumes have a chat demo and no metrics. I want to be the one who ships with evals: golden datasets, LLM-as-judge, regression tracking, and Docker/CI pipelines. This is my lever for standing out.

RAG metrics: faithfulness, relevance, context precision. I use these to measure before I optimize.

Traces, cost, latency. The observability I need to debug like I debug a React app.

MLOps practices: tests, CI/CD, production discipline. This is my engineering backbone.

Docker — Get Started

Multi-stage builds and slim images. I already use containers; here I make them sharp.

buildWire evals into my RAG project and publish before/after scores. This is the strongest 'scholarly' proof I can add to a portfolio.

P4 · MONTH 4

Agents

Agents feel like a state machine with tool calls, which is already close to my Redux and async-flow comfort zone. I am mapping nodes to reducers, state to store, and tool calls to side effects. LangGraph is my on-ramp here.

LangChain Academy

Official free LangGraph and agent courses. I work through them while building my own project.

Stateful agents: nodes, edges, checkpointing. Docs first, tutorials second.

DL.AI — Agent courses

Agentic RAG, memory, smolagents, human-in-the-loop. I pick what fits my project.

paid / audit↗

Generative AI with LLMs (DL.AI + AWS)

I audited the core content; the certificate is optional.

build · project 2An AI code-review assistant on LangGraph + a React dashboard. This is where my UI strength becomes a moat.

P5 · MONTH 5

Cloud + deployment + scale

I want to know enough to ship and argue about cost. I pick Azure first because it dominates Polish enterprise, and AWS second because it dominates product/US companies. I am not trying to become a DevOps engineer — I am trying to own the full loop.

Microsoft AI-102 (Azure AI Engineer)

Free and updated. This is my target for the Polish enterprise market.

AWS Bedrock docs

My parallel track for product companies and US-facing roles.

Designing ML Systems (Huyen)

Production architecture, monitoring, CI/CD. I read it as a complement to the AI Engineering book.

build · project 3 (flagship)A multi-agent system with real-time visualisation. Frontend-heavy execution is where 90% of AI engineers are weaker than I am.

P6 · MONTH 6

Portfolio + hiring + public track

My output target is three shipped frontend+AI projects, 5 written articles, and open-source work with real adoption. This is both a hiring portfolio and proof of original work. I am not just studying; I am publishing evidence.

alexisinwork/ai_engineering

My own repo of AI engineering notes, experiments, and reference code. I update it as I learn.

alexisinwork/free_AI_courses

A curated list of free AI courses I actually took or plan to take.

Raschka — Build an LLM from Scratch (repo)

Optional deep dive. I want to understand the internals, not just the API.

Optional, for when I learn better by building before I read the theory.

Optional deep learning foundation. I treat it as a reference, not a must-do.

shipPolish the 3 projects, publish the articles, and grow open-source adoption. Then I interview and apply to a short, targeted list of companies.

OUT · MONTH 6+

Endpoint: frontend + AI, shipped in public

My target is a working frontend+AI skill stack, a portfolio that proves it, and a public track of articles and open-source work. From there I am looking for a strong product-engineering role — either in a company that uses AI heavily, or a relocation-track opportunity that values the hybrid skill set.

How to read this map: P0–P6 is the exact 6-month track I am running. I am not following every resource in order; I am using them as references while I build the projects listed in each phase.

My frame: AI engineering is the engine, not a side hustle. Content and YouTube are background noise unless they directly support a project or article. I keep that time-boxed so it never replaces the actual skill building.

This page's performance

UI → AI roadmap · full page load

LCP: —
CLS: 0.000
INP: —
FCP: —
TTFB: —
RENDER: —

Measured live in your browser for this page only — observers reset on every route change. Full diagnostics →