# From Zero to AI Engineer in 2026: 12 Months, 5 Production-Level Projects, $200K+ Job Offer

> Source: https://sukruyusufkaya.com/en/blog/sifirdan-ai-engineer-yol-haritasi-12-ay-5-production-project-2026
> Updated: 2026-05-27T19:47:33.332Z
> Type: blog
> Category: yapay-zeka
**TLDR:** A concrete roadmap to land a global remote AI Engineer position from zero in 12 months: 5 production-level projects, GitHub portfolio + blog strategy, $200K+ offer. Karpathy, Raschka, 3Blue1Brown, Andrew Ng curriculum; HuggingFace + LangChain + Anthropic Academy free programs; Turkish alternatives; case study (14-month timeline); and interview strategy for top offers.

<tldr data-summary="[&quot;Going from zero to AI Engineer in 12 months is possible — but requires 25-35 hours/week of consistent work, 5 production-level projects, and active GitHub + blog + LinkedIn presence.&quot;,&quot;Month 1-2: math fundamentals (linear algebra, probability, calculus) + Python; Month 3: NumPy/Pandas/SQL; Month 4: Scikit-learn; Month 5-6: PyTorch + Deep Learning; Month 7-8: LLM stack + RAG; Month 9-10: Agents + LLMOps; Month 11-12: 5 projects + portfolio + interviews.&quot;,&quot;5 production projects: (1) Turkish LLM chatbot, (2) Enterprise RAG, (3) Multi-agent system, (4) Fine-tuned LLM, (5) End-to-end MLOps pipeline. Each marketed through GitHub + blog + LinkedIn.&quot;,&quot;Free resources are enough: Karpathy YouTube series (Zero to Hero), Raschka book/course, Andrew Ng DeepLearning.AI, HuggingFace courses, LangChain Academy, Anthropic Academy; total cost $0-300.&quot;,&quot;Interview prep: 150-300 LeetCode problems (must), Chip Huyen ML System Design book, ability to story-tell your production projects; average 6-10 interviews before first offer.&quot;,&quot;Case: someone who went from zero to AI Engineer in 14 months in 2025 landed Cohere Mid-level at $235K total comp — replicable structure.&quot;]" data-one-line="12 months of consistent work + 5 production-level projects + active portfolio marketing is the proven formula for going from zero to a $200K+ remote AI Engineer offer in 2026."></tldr>

## 1. Introduction: Is It Still Possible to Become an AI Engineer From Zero in 2026?

Short answer: **yes, but with more discipline than in 2022**. As of 2026, AI Engineer entry bar has risen because:

1. Demand exploded (+74% YoY) but supply also grew (15M+ developers using LLMs).
2. Companies now filter out profiles **without production experience**.
3. ChatGPT + Cursor + Claude Code accelerate entry-level skip — but to **show this as your own achievement**, you need real production projects.

<definition-box data-term="AI Engineer (Applied)" data-definition="A software specialist who connects the modern AI stack (LLM, RAG, agents, embeddings, vector DBs, LLMOps) to production-grade products. Differs from classical ML Engineer in building the application layer using pre-trained foundation models rather than training from scratch — system design, observability, and scalability are first-class concerns." data-also="LLM Engineer, GenAI Engineer, Applied AI Engineer" data-wikidata="Q124728960"></definition-box>

This guide presents a **12-month** consistent roadmap: with 3-5 hours/day (25-35 hours/week) of discipline, producing 5 production-level projects, reaching a **global remote** or **Türkiye-premium** AI Engineer position. No surprises — just **discipline + evidence**.

<stat-callout data-value="14 months" data-context="In 2025, an engineer profile starting from zero landed a Cohere Mid-level AI Engineer offer in" data-outcome="14 months — 85% of the journey matches this guide; 14 months of consistent work, 5 production-quality projects, and active GitHub + blog strategy." data-source="{&quot;label&quot;:&quot;Internal Case Study, 2025&quot;,&quot;url&quot;:&quot;https://sukruyusufkaya.com/blog/sifirdan-ai-engineer-yol-haritasi-12-ay-5-production-project-2026&quot;,&quot;date&quot;:&quot;2025-12&quot;}"></stat-callout>

### Who Is This Guide For?

- **Software developers:** Frontend/Backend/Mobile experienced, wanting to transition to AI Engineer.
- **New graduates:** Computer/Industrial/Math grads aiming for their first role.
- **Career changers:** From other sectors (finance, biology, physics) aiming at AI.
- **ML Engineers:** Evolving from classical ML to modern LLM-centric AI Engineering.

If you have no programming experience: this guide takes **22-26 months** for you (programming foundation + AI roadmap). The 12-month version assumes "knows how to code but doesn't know AI" starting point.

## 2. 12-Month Roadmap: Detailed Anatomy

### Month 1-2: Math Foundations + Python Polish

**Goal.** Solidify linear algebra, calculus, probability/statistics and AI-specific Python features.

#### Math (50 hours/month)

- **Linear Algebra:** Vectors/matrices, eigenvalue/eigenvector, SVD, projection.
- **Probability:** Bayes' theorem, expectation, variance, distributions (Gaussian, Bernoulli).
- **Calculus:** Derivative, partial derivative, chain rule (critical for backprop), gradient.

**Resources:**
- **3Blue1Brown** YouTube: "Essence of Linear Algebra" (~3 hours) — indispensable.
- **3Blue1Brown** "Essence of Calculus".
- **MIT OCW 18.06** Linear Algebra (Gilbert Strang) — for depth.
- **Imperial College Mathematics for ML** Coursera (free audit).

#### Python (40 hours/month)

- f-string, list/dict comprehension, generator, decorator, type hints.
- async/await, threading basics.
- venv, poetry, uv.

**Resources:**
- **Fluent Python** (Ramalho).
- **Sebastian Raschka** YouTube.
- **Real Python**.

### Month 3: NumPy + Pandas + Matplotlib + SQL

**Goal.** Make working with data your native skill.

- **NumPy:** ndarray, broadcasting, vectorization.
- **Pandas:** DataFrame, groupby, merge, pivot, time-series.
- **Matplotlib + Seaborn + Plotly:** Visualization.
- **SQL:** JOIN, window functions, CTE, indexing.

**Resources:** Python for Data Analysis (Wes McKinney), Kaggle Learn Pandas + SQL, DataCamp SQL.

### Month 4: Scikit-learn + Classical ML

**Goal.** Understand classical ML algorithms; master scikit-learn API.

- **Supervised:** Linear/Logistic, Decision Tree, Random Forest, XGBoost, SVM.
- **Unsupervised:** K-Means, DBSCAN, PCA, hierarchical clustering.
- **Pipeline:** train/test split, cross-validation, feature engineering, hyperparameter tuning, metrics.

**Resources:** Hands-On ML (Géron Ch 1-9), Andrew Ng Machine Learning Specialization, Raschka book.

#### Mini Project 1 (End of Month 4)

**Project:** Turkish Twitter sentiment classifier (classical ML: TF-IDF + Logistic Regression). GitHub + README + Notebook + Streamlit deploy.

### Month 5-6: PyTorch + Deep Learning

**Goal.** Understand neural networks; master PyTorch as your single tool.

- **PyTorch fundamentals:** Tensor, autograd, Module, optimizer, DataLoader.
- **CNN:** Convolution, pooling, ResNet, transfer learning.
- **RNN/LSTM:** Sequence models (rarely used now but important to understand).
- **Transformer:** Self-attention, multi-head, encoder/decoder, BERT/GPT.

**Resources:**
- **Andrej Karpathy** YouTube "Neural Networks: Zero to Hero" — 9-video series (~20 hours). **Indispensable** — building backprop from scratch is the fastest way to learn.
- **Karpathy** "Let's build GPT from scratch".
- **Deep Learning with PyTorch** (Eli Stevens).
- **Jay Alammar** "Illustrated Transformer".
- **DeepLearning.AI Deep Learning Specialization** Coursera.

#### Mini Project 2 (End of Month 6)

**Project:** CNN + transfer learning Turkish food image classifier (Foody Turkish Cuisine dataset). Deploy to Hugging Face Spaces.

### Month 7: LLM Stack — HuggingFace, Prompting, Fine-tuning Basics

**Goal.** Actually use the modern LLM ecosystem.

- **HuggingFace Transformers:** Pipeline, AutoModel, AutoTokenizer.
- **Prompting:** Few-shot, CoT, ReAct, system prompts.
- **APIs:** OpenAI, Anthropic, Cohere, Mistral, OpenRouter.
- **Fine-tuning basics:** LoRA, QLoRA, PEFT.

**Resources:** HuggingFace NLP Course, HuggingFace LLM Course, DeepLearning.AI Prompt Engineering, Anthropic Academy, OpenAI Cookbook.

### Month 8: RAG Architecture + Vector DB + Embeddings

**Goal.** Build a production-quality RAG system.

- **Embedding models:** BGE-M3, OpenAI text-embedding-3, Cohere embed-multilingual-v3.
- **Vector DBs:** Qdrant, Weaviate, Pinecone, pgvector, Milvus.
- **Chunking strategies:** Fixed, semantic, structural, hybrid.
- **Hybrid search:** BM25 + vector + Reciprocal Rank Fusion.
- **Re-ranking:** bge-reranker, Cohere rerank.
- **Eval:** RAGAS, DeepEval, TruLens.

**Resources:** LangChain RAG Cookbook, LlamaIndex Documentation, Pinecone Learning Center, OpenAI RAG Best Practices.

### Month 9: AI Agents + LangChain/LangGraph + MCP

**Goal.** Build multi-step autonomous systems.

- **Agent paradigms:** ReAct, Plan-and-Execute, multi-agent.
- **LangChain + LangGraph:** State machine, tool calling, memory.
- **MCP (Model Context Protocol):** Anthropic's 2024 open standard; tool + resource + prompt integration.
- **OpenAI Function Calling** + **Anthropic Tool Use** + **Gemini Function Calling**.
- **Smol agents** (HuggingFace).

**Resources:** LangChain Academy, DeepLearning.AI AI Agents in LangGraph (free), Anthropic MCP Documentation, HuggingFace Agents Course.

### Month 10: LLMOps + Observability + Evaluation

**Goal.** Learn to "put AI systems into production."

- **Observability:** Langfuse, Helicone, Arize Phoenix, LangSmith.
- **Eval harness:** RAGAS, DeepEval, custom metrics.
- **Cost tracking:** Token usage, model cost, optimization.
- **Versioning:** Prompt versioning, dataset, model.
- **CI/CD:** GitHub Actions for model deploy, eval regression.
- **Monitoring:** Latency, error rate, eval score drop alerts.

**Resources:** Chip Huyen Designing ML Systems (Ch 8-11), Eugene Yan blog, Hamel Husain blog, Langfuse Documentation.

### Month 11-12: 5 Production-Level Projects + Portfolio + Interview Prep

The last 2 months are **execution** — learning is done, building begins. Turn everything you learned into 5 portfolio projects, do GitHub + blog + LinkedIn marketing, start interviews.

## 3. 5 Production-Level Projects — Detailed

These 5 projects cover every skill signal a recruiter looks for in an AI Engineer: backend design, modern AI stack, production readiness, eval, observability, documentation.

### Project 1: Turkish-capable LLM-Powered Chatbot

**Concept.** A production-quality chatbot for Turkish-speaking users on OpenAI or Claude API — deployed.

**Technical requirements.**
- **Backend:** FastAPI (Python) or Next.js API routes.
- **LLM:** Claude Sonnet 4.7 or GPT-5; fallback Llama 4 (self-hosted).
- **Session management:** Redis or PostgreSQL.
- **Rate limiting:** Token bucket, per-user.
- **Streaming:** SSE or WebSocket.
- **Observability:** Langfuse integration; every conversation logged.
- **Frontend:** Next.js + shadcn/ui or Streamlit.
- **Deploy:** Vercel or Railway.

**Blog post:** "Shipping a Turkish Chatbot to Production: 12 Decisions and 5 Mistakes."

**LinkedIn:** Build-in-public strategy — 3 weeks of progress updates.

### Project 2: Enterprise RAG (PDF Ingestion + Hybrid Search + Eval Harness)

**Concept.** A RAG system that answers questions against a company's PDF + Word + Notion documents, with citations.

**Technical requirements.**
- **Ingestion:** PyMuPDF (PDF), python-docx (Word), Notion API.
- **Chunking:** Structural + 15% overlap (LangChain RecursiveCharacterTextSplitter).
- **Embedding:** BGE-M3 (self-hosted) or OpenAI text-embedding-3-large.
- **Vector DB:** Qdrant (Docker container) or pgvector.
- **Hybrid search:** Qdrant sparse + dense fusion.
- **Re-ranking:** bge-reranker-v2-m3.
- **LLM:** Claude Opus 4.7 (1M context — for long docs).
- **Eval:** RAGAS (faithfulness, context precision, context recall, answer relevance).
- **UI:** Streamlit or Next.js.

**Blog post:** "I Shipped a RAG System to Production: How I Raised Faithfulness from 0.68 to 0.91."

### Project 3: Multi-Agent System (LangGraph + MCP)

**Concept.** A system that solves a complex task (e.g., "write an article about this company" or "deep research on this topic") via multiple agents.

**Technical requirements.**
- **Framework:** LangGraph (state machine).
- **Agents:** (a) Planner, (b) Researcher, (c) Writer, (d) Editor, (e) Critic.
- **Tools:** Web search (Brave Search API), document retrieval, code execution (E2B).
- **MCP integration:** Write one MCP server and call tools via it.
- **State persistence:** Redis or SQLite.
- **Observability:** LangSmith + Langfuse.
- **Eval:** Task completion rate, average turn count, cost.

**Key value:** MCP became industry standard in 2026; an MCP server in your portfolio is a strong signal.

**Blog post:** "I Built an Autonomous Research System with 5 Agents: 47 Turns, $0.83 Cost, 3 Key Lessons."

### Project 4: Fine-tuned LLM (LoRA + Dataset + Deploy)

**Concept.** Fine-tune a base model (Llama 3.1 8B, Mistral 7B, or Qwen 2.5 7B) for your own domain (e.g., Turkish healthcare assistant, legal assistant, customer service).

**Technical requirements.**
- **Base model:** Llama 3.1 8B / Mistral 7B / Qwen 2.5 7B.
- **Dataset:** At least 5,000 examples (instruction-tuned format). For Turkish: self-curated or HuggingFace TR datasets.
- **Fine-tuning:** LoRA + PEFT (Unsloth library — 2-5x faster).
- **Compute:** Google Colab Pro+ (~$50/mo) or RunPod (~$0.50/hr A100).
- **Eval:** Custom eval set (50+ Q) + baseline vs fine-tuned comparison.
- **Deploy:** vLLM or Ollama; HuggingFace Spaces (free tier).

**Blog post:** "I Fine-tuned Llama 3.1 8B for Turkish with LoRA: 5,000 Examples, $80 Compute, 32% Eval Improvement."

### Project 5: End-to-End MLOps (CI/CD + Monitoring + Alerting)

**Concept.** Wrap one of the above projects (say the RAG) in a full MLOps pipeline.

**Technical requirements.**
- **CI/CD:** GitHub Actions — push triggers test, eval, deploy.
- **Container:** Docker + docker-compose; multi-stage build.
- **Orchestration:** Kubernetes (k3s or minikube) — optional but bonus.
- **Monitoring:** Prometheus + Grafana or Datadog free tier.
- **Logging:** Structured logging (JSON), Loki or CloudWatch.
- **Alerting:** Latency > 5s or eval score < 0.85 alerts (PagerDuty free or Discord webhook).
- **Cost monitoring:** Token usage dashboard, anomaly detection.
- **Versioning:** MLflow or Weights & Biases for model + prompt + dataset versioning.

**Blog post:** "Keeping an AI System in Production Solo: 6 Months, 14 Incidents, $230 Cloud Bill."

## 4. Comparison: This Roadmap vs roadmap.sh

<comparison-table data-caption="This Roadmap vs roadmap.sh AI Engineer (2026)" data-headers="[&quot;Dimension&quot;,&quot;This Guide&quot;,&quot;roadmap.sh&quot;]" data-rows="[{&quot;feature&quot;:&quot;Duration&quot;,&quot;values&quot;:[&quot;12 months&quot;,&quot;Unclear (6-18 mo)&quot;]},{&quot;feature&quot;:&quot;Structure&quot;,&quot;values&quot;:[&quot;Monthly modular plan&quot;,&quot;Branching skill tree&quot;]},{&quot;feature&quot;:&quot;Project Plan&quot;,&quot;values&quot;:[&quot;5 detailed production projects&quot;,&quot;Abstract topic list&quot;]},{&quot;feature&quot;:&quot;Turkish Resources&quot;,&quot;values&quot;:[&quot;Yes (Karpathy + TR alt)&quot;,&quot;No&quot;]},{&quot;feature&quot;:&quot;Interview Prep&quot;,&quot;values&quot;:[&quot;Included&quot;,&quot;No&quot;]},{&quot;feature&quot;:&quot;Tax/Remote&quot;,&quot;values&quot;:[&quot;Türkiye-focused&quot;,&quot;No&quot;]},{&quot;feature&quot;:&quot;Cost&quot;,&quot;values&quot;:[&quot;$0-300 total&quot;,&quot;$0&quot;]},{&quot;feature&quot;:&quot;Mentor Support&quot;,&quot;values&quot;:[&quot;Community + blog&quot;,&quot;No&quot;]}]"></comparison-table>

**Practical advice:** Use roadmap.sh as a concept tree initially; switch to this guide's **12-month structure + 5-project package**. roadmap.sh tells you "what to learn," this guide tells you "how and in what order."

## 5. Resource Curriculum (Free + Paid)

### YouTube Channels

- **Andrej Karpathy** — Deep Learning + Transformer (vital).
- **3Blue1Brown** — Math (irreplaceable for intuition).
- **Sebastian Raschka** — PyTorch + ML.
- **Yannic Kilcher** — Paper review.
- **Two Minute Papers** — AI news.

### Coursera + DeepLearning.AI

- Andrew Ng Machine Learning Specialization.
- DeepLearning.AI Deep Learning Specialization.
- DeepLearning.AI NLP Specialization.
- DeepLearning.AI Generative AI with LLMs.

### HuggingFace + LangChain + Anthropic Academy (Free)

- HuggingFace NLP Course.
- HuggingFace LLM Course.
- HuggingFace Agents Course (2026).
- LangChain Academy.
- Anthropic Academy.

### Books (In Order)

1. **Hands-On Machine Learning** (Géron).
2. **Deep Learning** (Goodfellow, Bengio, Courville).
3. **Machine Learning with PyTorch and Scikit-Learn** (Raschka).
4. **Hands-On Large Language Models** (Alammar, Grootendorst).
5. **Designing Machine Learning Systems** (Chip Huyen).
6. **AI Engineering** (Chip Huyen, 2025).

### Certification ROI

<comparison-table data-caption="Certification ROI for AI Engineer (2026)" data-headers="[&quot;Certification&quot;,&quot;Cost&quot;,&quot;Duration&quot;,&quot;ROI&quot;]" data-rows="[{&quot;feature&quot;:&quot;AWS Certified ML Specialty&quot;,&quot;values&quot;:[&quot;$300&quot;,&quot;3 mo&quot;,&quot;Medium&quot;]},{&quot;feature&quot;:&quot;Azure AI Engineer (AI-102)&quot;,&quot;values&quot;:[&quot;$165&quot;,&quot;2 mo&quot;,&quot;Medium&quot;]},{&quot;feature&quot;:&quot;Google PMLE&quot;,&quot;values&quot;:[&quot;$200&quot;,&quot;3 mo&quot;,&quot;Medium-High&quot;]},{&quot;feature&quot;:&quot;NVIDIA DLI&quot;,&quot;values&quot;:[&quot;$0-90&quot;,&quot;Short&quot;,&quot;Low&quot;]},{&quot;feature&quot;:&quot;DeepLearning.AI Coursera&quot;,&quot;values&quot;:[&quot;$50/mo&quot;,&quot;4-6 mo&quot;,&quot;High (learning value)&quot;]},{&quot;feature&quot;:&quot;HuggingFace Cert&quot;,&quot;values&quot;:[&quot;$0&quot;,&quot;2-3 mo&quot;,&quot;Low-Medium&quot;]},{&quot;feature&quot;:&quot;Anthropic Academy&quot;,&quot;values&quot;:[&quot;$0&quot;,&quot;Short&quot;,&quot;High (for Claude)&quot;]}]"></comparison-table>

**Practical advice:** Certifications **do not replace a portfolio**. AWS ML / Azure AI / Google PMLE can be confidence-boosters during a career transition but are insufficient alone.

## 6. Türkiye-Specific Angle: Remote, Tax, Interview Strategy

### First Job: Türkiye Local vs Direct Global Remote

<comparison-table data-caption="First Job Strategy: Local vs Global Remote" data-headers="[&quot;Strategy&quot;,&quot;Duration&quot;,&quot;Salary&quot;,&quot;Pro&quot;,&quot;Con&quot;]" data-rows="[{&quot;feature&quot;:&quot;TR local junior&quot;,&quot;values&quot;:[&quot;6-8 weeks&quot;,&quot;35-55K TL/mo&quot;,&quot;Quick entry, mentorship&quot;,&quot;Low pay&quot;]},{&quot;feature&quot;:&quot;TR premium tech&quot;,&quot;values&quot;:[&quot;8-14 weeks&quot;,&quot;45-75K TL/mo&quot;,&quot;Modern stack&quot;,&quot;Selective&quot;]},{&quot;feature&quot;:&quot;Global remote Junior&quot;,&quot;values&quot;:[&quot;16-30 weeks&quot;,&quot;$2.5K-5K/mo&quot;,&quot;High pay&quot;,&quot;Very hard&quot;]},{&quot;feature&quot;:&quot;TR 2 years + global remote&quot;,&quot;values&quot;:[&quot;12-18 mo + 8-16 weeks&quot;,&quot;$5K-12K/mo&quot;,&quot;Optimal&quot;,&quot;Requires patience&quot;]}]"></comparison-table>

**Practical strategy:** **First 1.5-2 years in Türkiye** (local junior or premium tech), then **transition to global remote**. This route produces results faster than starting global remote from scratch.

### Tax Structure: Earning USD from Türkiye

- **Sole Proprietorship:** Optimal for < $80K/year — service export VAT 0%, income tax 15-37%.
- **LLC:** Optimal for $80K-$200K/year — effective 32% tax.
- **EOR (Deel, Remote.com):** Stable salary, simple, but 12-18% EOR fee.

Details: AI Engineer Salary Report guide.

### LinkedIn Profile Optimization (AI Engineer Specific)

5 keys:
1. **Headline:** "AI Engineer | RAG, Agents, LLMOps" — keywords.
2. **About:** 3 sentences — what you do, work on, what you've shipped.
3. **Featured:** Pin the 3 strongest projects (GitHub + blog + demo).
4. **Skills:** 30+ skills, 10 with 99+ endorsements.
5. **Public activity:** 1-2 posts/week (build-in-public).

## 7. Case Study: From Zero to Cohere Mid-Level in 14 Months

**Profile.** In 2024, a 4-year Backend Developer (TR fintech), CS bachelor, zero ML/AI knowledge. Goal: From zero to AI Engineer → Global remote.

**Journey.**

- **Month 1-2.** 3Blue1Brown linear algebra + calculus (8 hours/week). Python polish.
- **Month 3.** Kaggle Pandas + SQL courses. First Kaggle competition (top 30%).
- **Month 4.** Géron book Ch 1-9. Mini Project 1: Turkish sentiment classifier.
- **Month 5-6.** Karpathy "Zero to Hero" full series (15 hours effective video + 35 hours practice). PyTorch + transfer learning. Mini Project 2: food classifier.
- **Month 7.** HuggingFace NLP Course + DeepLearning.AI Prompt Engineering. First LLM API calls.
- **Month 8.** LangChain + Qdrant first RAG MVP. Blog post (Medium, 1,200 words).
- **Month 9.** LangGraph multi-agent prototype. Read Anthropic MCP docs.
- **Month 10.** LLMOps deep-dive: Langfuse, RAGAS. Production-grade RAG eval harness.
- **Month 11-12.** Project 1 (chatbot), Project 2 (RAG), Project 3 (multi-agent), Project 4 (LoRA fine-tune Llama 3.1 8B), Project 5 (MLOps pipeline). Each + blog + LinkedIn post.
- **Month 13.** 60+ applications (Cohere, Mistral, Hugging Face, Pinecone, LangChain, Anthropic, Series B/C startups). 12 responses, 4 first-round, 2 final-round.
- **Month 14.** Cohere Mid-level AI Engineer offer: **$235K base + $40K bonus + 0.2% equity**. EOR via Deel.

**Total investment.**
- Hours: ~1,450 hours (14 months × ~104 hours/month).
- Money: $280 (Colab Pro $50, RunPod compute $80, domain + hosting $30, books $120).

**Key takeaways.**
1. **Consistency** > sprint. 25-30 hours/week for 14 months > one-shot 60 hours.
2. **Build in public** is highly effective. LinkedIn progress posts brought 4 inbound recruiters.
3. **5 production projects** gave material to discuss during interviews.
4. **Karpathy "Zero to Hero"** was the turning point. Building Transformer from scratch proved depth.

## 8. Risks and Traps

<callout-box data-variant="warning" data-title="Tutorial Hell">

The biggest trap of this guide: **constantly watching courses, never writing code**. Spending 200 hours of video and shipping 0 production project = zero career value. Rule: **write code for at least 1 hour after every video**. Karpathy "Zero to Hero" is 15 hours of video but requires 35 hours of practice.

</callout-box>

<callout-box data-variant="warning" data-title="Starting Interviews Too Early">

Impatiently starting interviews at Month 6-7 — result: rejections, demotivation. Do not start interviews before Month 11-12; portfolio + storytelling insufficient.

</callout-box>

<callout-box data-variant="warning" data-title="Over-attaching to One Framework">

Treating LangChain as a deity — when the framework shifts (LangGraph, LlamaIndex, writing your own agent library), panic. **Learn the primitives below the framework**: prompt template, tool calling, state management — these transfer.

</callout-box>

<callout-box data-variant="warning" data-title="Buying into Fine-tuning Hype">

"I'll fine-tune everything" trap. Reality: **RAG + good prompting** beats fine-tuning in 80% of cases. Learn fine-tuning basics in Month 8, apply in Project 4 — but in AI Engineer roles this is not the daily job.

</callout-box>

<callout-box data-variant="warning" data-title="Skipping LeetCode">

Interview prep is impossible without LeetCode. For a global remote AI Engineer offer, solving 150-300 LeetCode problems is **minimum**. Don't push it to the end — start in Month 9-10, solve 5-8 problems/week, hit 200-300 by Month 12.

</callout-box>

## 9. Frequently Asked Questions

<callout-box data-variant="answer" data-title="I have no software experience — can I follow this roadmap?">

**Not directly** — this guide assumes a "knows how to code but doesn't know AI" starting point. If you have zero coding: spend **6-8 months** on Python + software fundamentals (Git, Linux, SQL, REST APIs) first. CS50 (Harvard) + Automate the Boring Stuff (Sweigart) are good starts. Then come to this guide. Total: ~20-22 months.

</callout-box>

<callout-box data-variant="answer" data-title="How many hours per week should I invest?">

**Minimum 25 hours/week**. If you work full-time: evenings + weekends combo. If unemployed or part-time: 40-50 hours/week — you can shorten this roadmap to 8-9 months. Less than 15 hours/week: roadmap stretches to 18-24 months with high motivation-loss risk.

</callout-box>

<callout-box data-variant="answer" data-title="Can I become an AI Engineer without a PhD?">

**Absolutely yes.** AI Engineer is "Applied AI" — building production systems. PhD is generally needed for **research scientist** roles (OpenAI MTS, Anthropic Research, DeepMind Research). For Applied AI Engineer, a 4-year degree + portfolio + 5 production projects suffices.

</callout-box>

<callout-box data-variant="answer" data-title="Which LLM API should I use — OpenAI, Anthropic, or Google?">

**Use all three** — real production systems often have multi-LLM fallback. Start: **OpenAI** (widest ecosystem), **Anthropic Claude** (best reasoning + tool use), **Google Gemini** (cheap + multimodal). For cost: aggregators like OpenRouter or Together.ai (50+ models, single API).

</callout-box>

<callout-box data-variant="answer" data-title="Which framework: LangChain, LlamaIndex, or self-roll?">

**Junior-Mid:** LangChain — widest ecosystem, most community resources. **Mid-Senior:** LangGraph (for state management) or try self-rolling (more control). **Senior+:** typically write your own minimalist library because framework limits become apparent. LlamaIndex is specifically good for RAG. Don't lock into one framework — primitives (prompt template, tool calling) are the same everywhere.

</callout-box>

<callout-box data-variant="answer" data-title="How many LeetCode problems should I solve?">

**Minimum 150** (Top Interview Questions Easy + Medium); **target 250-300** (Medium-heavy + some Hard). In 2026, AI Engineer interviews include classic DSA + ML system design + LLM system design. LeetCode alone isn't enough, but you can't go without it. Strategy: NeetCode 150 (curated list) → Blind 75 → company-specific (Cohere, Anthropic for OOP-heavy).

</callout-box>

<callout-box data-variant="answer" data-title="What helper tools should I use while following this guide?">

**Coding:** Cursor (with Claude Sonnet 4.5+) or Claude Code. 3-5x acceleration. **Notes:** Obsidian + daily notes (critical for retention). **Spaced repetition:** Anki (math formulas + algorithm patterns). **Versioning:** Git (commit daily) + GitHub. **Tracking:** A Notion table — daily hours, topic, completed practice. For discipline.

</callout-box>

<callout-box data-variant="answer" data-title="What if I haven't found a job after 14 months?">

**Re-evaluate.** Common reasons: **(1)** Weak portfolio — projects shallow or missing. **(2)** Low GitHub commit history — recruiters check. **(3)** Insufficient interview prep — LeetCode + ML system design. **(4)** Passive LinkedIn — no outreach or build-in-public. **(5)** Wrong target market — global remote junior is very hard; open up to Türkiye-local roles. Get a mentor (paid or volunteer) for portfolio + LinkedIn review.

</callout-box>

## 10. Next Steps

To put this roadmap into action:

1. **Week 1-2:** Finish 3Blue1Brown Linear Algebra (3 hours video + 6 hours practice). Polish Python with "Fluent Python" Ch 1-3.
2. **Week 3-4:** Finish Karpathy "micrograd" + "makemore". Build backprop from scratch.
3. **Week 5-8:** First Kaggle competition (Titanic + House Prices). Upload Notebook + README to GitHub.
4. **Monthly check-in:** Notion table for weekly progress.
5. **3-month milestone:** Mini project + blog post + LinkedIn share.

**If you need a mentor**, reach out via the site's contact form — structured mentorship packages for the 12-month goal are available.

<references-list data-items="[{&quot;title&quot;:&quot;Andrej Karpathy - Neural Networks: Zero to Hero&quot;,&quot;url&quot;:&quot;https://www.youtube.com/playlist?list=PLAqhIrjkxbuWI23v9cThsA9GvCAUhRvKZ&quot;,&quot;author&quot;:&quot;Andrej Karpathy&quot;,&quot;publishedAt&quot;:&quot;2022-2025&quot;,&quot;publisher&quot;:&quot;YouTube&quot;},{&quot;title&quot;:&quot;3Blue1Brown - Essence of Linear Algebra&quot;,&quot;url&quot;:&quot;https://www.youtube.com/playlist?list=PLZHQObOWTQDPD3MizzM2xVFitgF8hE_ab&quot;,&quot;author&quot;:&quot;Grant Sanderson&quot;,&quot;publishedAt&quot;:&quot;2016-2026&quot;,&quot;publisher&quot;:&quot;YouTube&quot;},{&quot;title&quot;:&quot;Hands-On Machine Learning - 3rd Edition&quot;,&quot;url&quot;:&quot;https://www.oreilly.com/library/view/hands-on-machine-learning/9781098125967/&quot;,&quot;author&quot;:&quot;Aurélien Géron&quot;,&quot;publishedAt&quot;:&quot;2023&quot;,&quot;publisher&quot;:&quot;O&apos;Reilly&quot;},{&quot;title&quot;:&quot;Sebastian Raschka - Machine Learning with PyTorch and Scikit-Learn&quot;,&quot;url&quot;:&quot;https://sebastianraschka.com/books/&quot;,&quot;author&quot;:&quot;Sebastian Raschka&quot;,&quot;publishedAt&quot;:&quot;2022&quot;,&quot;publisher&quot;:&quot;Packt&quot;},{&quot;title&quot;:&quot;Hands-On Large Language Models&quot;,&quot;url&quot;:&quot;https://www.llm-book.com/&quot;,&quot;author&quot;:&quot;Jay Alammar, Maarten Grootendorst&quot;,&quot;publishedAt&quot;:&quot;2024&quot;,&quot;publisher&quot;:&quot;O&apos;Reilly&quot;},{&quot;title&quot;:&quot;Designing Machine Learning Systems&quot;,&quot;url&quot;:&quot;https://huyenchip.com/ml-interviews-book/&quot;,&quot;author&quot;:&quot;Chip Huyen&quot;,&quot;publishedAt&quot;:&quot;2022&quot;,&quot;publisher&quot;:&quot;O&apos;Reilly&quot;},{&quot;title&quot;:&quot;AI Engineering&quot;,&quot;url&quot;:&quot;https://www.oreilly.com/library/view/ai-engineering/9781098166298/&quot;,&quot;author&quot;:&quot;Chip Huyen&quot;,&quot;publishedAt&quot;:&quot;2025&quot;,&quot;publisher&quot;:&quot;O&apos;Reilly&quot;},{&quot;title&quot;:&quot;Andrew Ng - Machine Learning Specialization&quot;,&quot;url&quot;:&quot;https://www.coursera.org/specializations/machine-learning-introduction&quot;,&quot;author&quot;:&quot;Andrew Ng&quot;,&quot;publishedAt&quot;:&quot;2022-2026&quot;,&quot;publisher&quot;:&quot;DeepLearning.AI&quot;},{&quot;title&quot;:&quot;DeepLearning.AI Deep Learning Specialization&quot;,&quot;url&quot;:&quot;https://www.coursera.org/specializations/deep-learning&quot;,&quot;author&quot;:&quot;Andrew Ng&quot;,&quot;publishedAt&quot;:&quot;2017-2026&quot;,&quot;publisher&quot;:&quot;DeepLearning.AI&quot;},{&quot;title&quot;:&quot;DeepLearning.AI Generative AI with LLMs&quot;,&quot;url&quot;:&quot;https://www.coursera.org/learn/generative-ai-with-llms&quot;,&quot;author&quot;:&quot;DeepLearning.AI + AWS&quot;,&quot;publishedAt&quot;:&quot;2023-2026&quot;,&quot;publisher&quot;:&quot;DeepLearning.AI&quot;},{&quot;title&quot;:&quot;HuggingFace NLP Course&quot;,&quot;url&quot;:&quot;https://huggingface.co/learn/nlp-course&quot;,&quot;author&quot;:&quot;Hugging Face&quot;,&quot;publishedAt&quot;:&quot;2022-2026&quot;,&quot;publisher&quot;:&quot;Hugging Face&quot;},{&quot;title&quot;:&quot;HuggingFace Agents Course&quot;,&quot;url&quot;:&quot;https://huggingface.co/learn/agents-course&quot;,&quot;author&quot;:&quot;Hugging Face&quot;,&quot;publishedAt&quot;:&quot;2025-2026&quot;,&quot;publisher&quot;:&quot;Hugging Face&quot;},{&quot;title&quot;:&quot;LangChain Academy&quot;,&quot;url&quot;:&quot;https://academy.langchain.com/&quot;,&quot;author&quot;:&quot;LangChain&quot;,&quot;publishedAt&quot;:&quot;2024-2026&quot;,&quot;publisher&quot;:&quot;LangChain&quot;},{&quot;title&quot;:&quot;Anthropic Academy + Claude Skills&quot;,&quot;url&quot;:&quot;https://www.anthropic.com/learn&quot;,&quot;author&quot;:&quot;Anthropic&quot;,&quot;publishedAt&quot;:&quot;2024-2026&quot;,&quot;publisher&quot;:&quot;Anthropic&quot;},{&quot;title&quot;:&quot;Anthropic MCP Documentation&quot;,&quot;url&quot;:&quot;https://modelcontextprotocol.io/&quot;,&quot;author&quot;:&quot;Anthropic&quot;,&quot;publishedAt&quot;:&quot;2024-2026&quot;,&quot;publisher&quot;:&quot;Anthropic&quot;},{&quot;title&quot;:&quot;roadmap.sh AI Engineer&quot;,&quot;url&quot;:&quot;https://roadmap.sh/ai-engineer&quot;,&quot;author&quot;:&quot;roadmap.sh&quot;,&quot;publishedAt&quot;:&quot;2024-2026&quot;,&quot;publisher&quot;:&quot;roadmap.sh&quot;},{&quot;title&quot;:&quot;Eugene Yan - Applied ML&quot;,&quot;url&quot;:&quot;https://eugeneyan.com/&quot;,&quot;author&quot;:&quot;Eugene Yan&quot;,&quot;publishedAt&quot;:&quot;2020-2026&quot;,&quot;publisher&quot;:&quot;Personal Blog&quot;},{&quot;title&quot;:&quot;Hamel Husain - LLM Evaluation&quot;,&quot;url&quot;:&quot;https://hamel.dev/&quot;,&quot;author&quot;:&quot;Hamel Husain&quot;,&quot;publishedAt&quot;:&quot;2023-2026&quot;,&quot;publisher&quot;:&quot;Personal Blog&quot;},{&quot;title&quot;:&quot;Yannic Kilcher YouTube&quot;,&quot;url&quot;:&quot;https://www.youtube.com/@YannicKilcher&quot;,&quot;author&quot;:&quot;Yannic Kilcher&quot;,&quot;publishedAt&quot;:&quot;2018-2026&quot;,&quot;publisher&quot;:&quot;YouTube&quot;},{&quot;title&quot;:&quot;Fluent Python - 2nd Edition&quot;,&quot;url&quot;:&quot;https://www.oreilly.com/library/view/fluent-python-2nd/9781492056348/&quot;,&quot;author&quot;:&quot;Luciano Ramalho&quot;,&quot;publishedAt&quot;:&quot;2022&quot;,&quot;publisher&quot;:&quot;O&apos;Reilly&quot;},{&quot;title&quot;:&quot;Python for Data Analysis - 3rd Edition&quot;,&quot;url&quot;:&quot;https://wesmckinney.com/book/&quot;,&quot;author&quot;:&quot;Wes McKinney&quot;,&quot;publishedAt&quot;:&quot;2022&quot;,&quot;publisher&quot;:&quot;O&apos;Reilly&quot;},{&quot;title&quot;:&quot;Deep Learning (Goodfellow Book)&quot;,&quot;url&quot;:&quot;https://www.deeplearningbook.org/&quot;,&quot;author&quot;:&quot;Goodfellow, Bengio, Courville&quot;,&quot;publishedAt&quot;:&quot;2016&quot;,&quot;publisher&quot;:&quot;MIT Press&quot;},{&quot;title&quot;:&quot;Jay Alammar - Illustrated Transformer&quot;,&quot;url&quot;:&quot;https://jalammar.github.io/illustrated-transformer/&quot;,&quot;author&quot;:&quot;Jay Alammar&quot;,&quot;publishedAt&quot;:&quot;2018-2026&quot;,&quot;publisher&quot;:&quot;Personal Blog&quot;},{&quot;title&quot;:&quot;NeetCode 150&quot;,&quot;url&quot;:&quot;https://neetcode.io/practice&quot;,&quot;author&quot;:&quot;NeetCode&quot;,&quot;publishedAt&quot;:&quot;2021-2026&quot;,&quot;publisher&quot;:&quot;NeetCode&quot;},{&quot;title&quot;:&quot;Chip Huyen - ML Interview Book&quot;,&quot;url&quot;:&quot;https://huyenchip.com/ml-interviews-book/&quot;,&quot;author&quot;:&quot;Chip Huyen&quot;,&quot;publishedAt&quot;:&quot;2023&quot;,&quot;publisher&quot;:&quot;Self-published&quot;}]"></references-list>

---

This is a living document; the AI ecosystem shifts every quarter, so the roadmap is **updated quarterly**.