Skip to content

README.md

Latest commit

 

History

History
281 lines (222 loc) · 12.8 KB

README.md

File metadata and controls

281 lines (222 loc) · 12.8 KB
Raw

AI Engineer Learning Track

"Build intelligent products that millions of people use every day."

AI Engineers build AI-powered applications and systems using Large Language Models (LLMs) and other foundation models. This is the fastest-growing engineering specialty in 2024-2025, and the barrier to entry is lower than you might think.

Track Overview

PHASE 1: LLM Foundations (1-2 months)
    Python → LLM APIs → Prompt Engineering → RAG Basics
         ↓
PHASE 2: Core AI Engineering (2-3 months)
    RAG Systems → LangChain → Agents → Vector Databases
         ↓
PHASE 3: Advanced AI Systems (3-5 months)
    Fine-tuning → Multi-agent → Evals → Production AI
         ↓
PHASE 4: Expert (Ongoing)
    Alignment → Research → Open Source Contributions

Prerequisites

  • Python proficiency (functions, classes, async/await basics)
  • Basic understanding of REST APIs
  • Git basics (clone, commit, push)
  • API key from OpenAI or Anthropic (free tier is fine)

Good news: This is the most accessible of the three tracks for Python developers. You can build real AI apps within your first week!

What Makes AI Engineering Different

Traditional Software AI Engineering
Deterministic output Probabilistic / non-deterministic output
Unit testable Requires custom evaluation frameworks
Bug = code error Bug = prompt issue, model hallucination, context problem
Deploy once, stable Models drift, need continuous evaluation
Clear correctness "Good enough" is often the standard

Skills You Will Build

Category Skills
LLM APIs OpenAI, Anthropic Claude, Google Gemini, Mistral, open-source
Frameworks LangChain, LlamaIndex, AutoGen, CrewAI
Vector Databases Chroma, Pinecone, Weaviate, Qdrant, pgvector
Embeddings OpenAI Embeddings, SentenceTransformers
Prompt Engineering Zero-shot, few-shot, CoT, ReAct, meta-prompting
RAG Naive RAG → Advanced RAG → GraphRAG
Agents Tool use, function calling, agent frameworks
Fine-tuning LoRA, QLoRA, DPO, RLHF basics
Evaluation RAGAS, DeepEval, LangSmith, custom evals
Deployment FastAPI, Modal, Hugging Face Spaces, Streamlit
Observability LangSmith, Arize Phoenix, Helicone
Responsible AI Bias detection, Fairness metrics, Explainability (XAI), Guardrails, Constitutional AI

Beginner Phase — LLM Foundations

Goal: Call LLM APIs confidently, understand prompt engineering, and build your first AI app.

Duration: 1-2 months (8-12 hrs/week)

Week Topic Resource Project
1 LLM fundamentals & APIs (OpenAI, Claude, Gemini) LLM APIs Guide Call 3 different LLM APIs
2-3 Prompt Engineering fundamentals Prompt Eng Guide Improve a bad prompt
4 Chatbot with memory Chatbot Guide Multi-turn chatbot
5-6 Basic RAG pipeline RAG Basics Q&A over your own docs
7-8 Deploy AI App (Streamlit + cloud) Deployment Guide Streamlit app + deploy

→ Start Beginner Phase

Intermediate Phase — Core AI Engineering

Goal: Build production-quality RAG systems, understand agents, and deploy AI features.

Duration: 2-3 months (10-15 hrs/week)

Week Topic Resource Project
1-2 LangChain deep dive LangChain Guide Build a chain
3-4 Advanced RAG techniques Advanced RAG Hybrid search + reranking
5-6 Vector databases Vector DB Guide Build semantic search
7-8 AI Agents Agents Guide Build tool-using agent
9-10 Function calling & tool use Tool Use Guide Agent with 5+ tools
11-12 LlamaIndex LlamaIndex Guide Complex document QA
13-14 AI Evaluation Eval Guide Build eval suite
15-16 Production AI Systems Production Guide Prod-ready AI service

→ Start Intermediate Phase

Advanced Phase — Expert AI Systems

Goal: Fine-tune LLMs, build multi-agent systems, and design production-grade AI architectures.

Duration: 3-5 months (12-15 hrs/week)

Week Topic Resource Project
1-3 LLM Fine-tuning (LoRA/QLoRA + Alignment) Fine-tuning Guide Fine-tune Llama 3
4-6 Multi-agent systems Multi-agent Guide Research + report agent
7-9 Advanced Retrieval techniques Advanced Retrieval Multi-strategy retrieval
10-12 AI Ethics, Safety & Guardrails Ethics & Safety Guide Implement a responsible AI toolkit (bias detection, guardrails)
13-15 GraphRAG & Knowledge Graphs GraphRAG Guide Knowledge graph RAG
16-18 Custom LLM Evaluation Advanced Evals Custom eval framework
19-21 AI Product Case Studies Case Studies (find and analyze 3 public post-mortems) Analyze + present

→ Start Advanced Phase

Projects

Beginner Projects

  • AI Chatbot with memory (OpenAI / Claude API)
  • Document Q&A System (basic RAG)
  • AI Writing Assistant (prompt engineering showcase)
  • YouTube Transcript Summarizer

Intermediate Projects

  • Personal Knowledge Base with RAG (your notes → searchable AI)
  • AI Research Assistant Agent (searches web + summarizes)
  • Multi-model AI API Gateway (route to different models)
  • Code Review Bot (GitHub Actions + LLM)

Advanced Projects

  • Fine-tuned LLM for a specific domain (legal, medical, etc.)
  • Multi-agent research system (AutoGen / CrewAI)
  • AI-powered SaaS product (idea → full product)
  • Custom Evaluation Framework for LLM outputs

→ See All Projects

Key Concepts You Must Understand

The AI Engineering Stack

┌──────────────────────────────────────────────────┐
│              AI Application Layer                 │
│  (Streamlit / FastAPI / Next.js / Slack bot)     │
├──────────────────────────────────────────────────┤
│              Orchestration Layer                  │
│  (LangChain / LlamaIndex / AutoGen / CrewAI)     │
├──────────────────────────────────────────────────┤
│              Memory & Storage Layer               │
│  (Vector DB / SQL / Redis / Knowledge Graph)     │
├──────────────────────────────────────────────────┤
│                 LLM Layer                         │
│  (OpenAI / Claude / Gemini / Llama / Mistral)   │
├──────────────────────────────────────────────────┤
│             Evaluation & Monitoring               │
│  (RAGAS / LangSmith / Arize / DeepEval)         │
└──────────────────────────────────────────────────┘

RAG Architecture

User Query
    ↓
Query Embedding
    ↓
Vector Similarity Search (Vector DB)
    ↓
Retrieved Chunks (Top-K)
    ↓
[Query + Context] → LLM
    ↓
Answer

Agent Loop (ReAct Pattern)

Thought → Action → Observation → Thought → ... → Final Answer

Skills Checklist

Beginner Level

  • Can call OpenAI and Anthropic APIs
  • Understand tokens, context windows, temperature
  • Can write effective prompts (zero-shot, few-shot, system prompts)
  • Built a basic chatbot with conversation history
  • Can build a basic RAG pipeline
  • Deployed one AI app (Streamlit, Hugging Face Spaces, etc.)

Intermediate Level

  • Built a production-quality RAG system with hybrid search
  • Built an AI agent that uses tools (web search, calculator, code execution)
  • Understand chunking strategies for RAG
  • Can evaluate RAG systems (faithfulness, relevance, completeness)
  • Know when to use LangChain vs LlamaIndex vs raw API calls
  • Can handle streaming responses in a web app

Advanced Level

  • Fine-tuned an open-source LLM with LoRA
  • Built and orchestrated a multi-agent system
  • Designed and ran a comprehensive LLM evaluation suite
  • Understand alignment basics (RLHF, DPO, Constitutional AI)
  • Can architect a production AI system with monitoring + guardrails
  • Can implement basic bias detection and fairness evaluation for a model
  • Contributed to an AI open-source project

Recommended Resources

Courses (Mostly Free)

Course Provider Duration
Neural Networks: Zero to Hero Andrej Karpathy (YouTube) 20 hrs
Prompt Engineering for Developers DeepLearning.AI 1-2 hrs
LangChain for LLM Apps DeepLearning.AI 1-2 hrs
Building RAG Agents with LLMs NVIDIA DLI 8 hrs
Generative AI for Beginners Microsoft (GitHub) 18 lessons
LangChain Academy LangChain Self-paced
Hugging Face NLP Course Hugging Face Self-paced

Certifications Worth Getting

  • IBM AI Engineering Professional Certificate (Coursera) — RAG, agents, LangChain end-to-end
  • IBM RAG and Agentic AI Professional Certificate (Coursera)
  • AWS Certified AI Practitioner
  • Microsoft Certified: Azure AI Engineer Associate

Essential Reading (Free)

Resource Link
Prompt Engineering Guide dair-ai/Prompt-Engineering-Guide
OpenAI Cookbook openai/openai-cookbook
LangChain Docs python.langchain.com
RAGAS Paper arXiv: 2309.15217
Attention Is All You Need arXiv: 1706.03762

AI Ethics & Responsible AI

Resource Link
Responsible AI Knowledge Base (GitHub) alexandrainst/responsible-ai
Google Research on Responsible AI research.google/teams/responsible-ai
AI Ethics: A Practical Guide for Responsible Use SBS Cyber
AI Ethics in Practice: Bias Detection & Fairness DEV Community
Actionable AI Ethics Kaggle Notebook
MIT Course: Ethics and Risks of AI MIT Professional Education
Skillsoft Course: AI Risk Management Skillsoft
Google Cloud: Responsible AI for Digital Leaders YouTube
Google Cloud: Applying AI Principles YouTube
Course: How to Use AI Responsibly in Business YouTube
Building Responsible AI Systems YouTube
Microsoft Responsible AI - Accountability YouTube
Paper: Against Explainable AI in Law arXiv: 2608.07452v1

Key Papers Every AI Engineer Should Read

  1. "Attention Is All You Need" (Transformer architecture)
  2. "Language Models are Few-Shot Learners" (GPT-3)
  3. "RLHF: Learning to summarize from human feedback"
  4. "Constitutional AI: Harmlessness from AI Feedback"
  5. "Retrieval-Augmented Generation for Knowledge-Intensive NLP"
  6. "LoRA: Low-Rank Adaptation of Large Language Models"

Back to: Main README | Role Comparison