AI Engineer Roadmap
This sheet is made for people who want to learn AI engineering for free in a well-organised and structured manner
Vectors, matrices, eigenvalues, matrix operations
Probability distributions, Bayes theorem, hypothesis testing
Derivatives, gradients, chain rule, partial derivatives
Variables, data types, control flow, loops, functions
Lists, tuples, dictionaries, sets, comprehensions
Classes, inheritance, polymorphism, encapsulation
Reading/writing files, error handling, context managers
NumPy, Pandas, Matplotlib, Seaborn basics
Data Structures from Scratch
beginnerImplement core data structures (Stack, Queue, Linked List) using Python classes.
- Python
- OOP
Commits, staging, branches, merging, rebasing, resolving conflicts, remotes, pull requests, .gitignore
SELECT, WHERE, JOINs, GROUP BY, aggregates, subqueries, window functions, schemas, indexes, querying from Python
Shell navigation, files and permissions, pipes and redirection, environment variables, processes, SSH, shell scripting
Compute, storage, networking, IAM, managed services, serverless, environment variables and secrets, deploying a container to the cloud
UI components, layouts, state management
Building ML/data visualization apps
Interactive Data Dashboard
beginnerBuild a sales or stock market dashboard with filters and interactive charts.
- Streamlit
- Pandas
- Plotly
REST APIs, path operations, request/response models
Request validation, response models, custom validators
async/await, asynchronous endpoints, background tasks
SQLAlchemy, database connections, ORM basics
JWT tokens, OAuth2, API keys, security best practices
Containerization, Docker Compose, production deployment
Task Management API
intermediateCreate a full CRUD (Create, Read, Update, Delete) REST API for a To-Do application.
- FastAPI
- Pydantic
- SQLite
Supervised vs Unsupervised learning, bias-variance tradeoff
Linear Regression, Ridge, Lasso, Polynomial Regression
Logistic Regression, Decision Trees, Random Forest, SVM, KNN
Feature scaling, encoding, selection, transformation
Accuracy, Precision, Recall, F1-Score, ROC-AUC, Confusion Matrix
Titanic Survival Predictor
beginnerBuild a classification model to predict passenger survival on the Titanic dataset.
- Scikit-learn
- Pandas
- Logistic Regression
House Price Regression
beginnerPredict housing prices using regression techniques and feature engineering.
- Scikit-learn
- Linear Regression
- Matplotlib
Iris Flower Classification
beginnerThe 'Hello World' of ML: Classify iris flowers based on sepal and petal measurements.
- Scikit-learn
- Decision Trees
Tokenization, stemming, lemmatization, stop words removal
Bag of Words, TF-IDF, N-grams
Word2Vec, GloVe, FastText
Sentiment Analysis, Text Classification, Named Entity Recognition
BLEU, ROUGE, Perplexity, F1-Score for NLP
SMS Spam Classifier
intermediateBuild a text classifier to detect spam messages using Bag-of-Words and TF-IDF.
- NLTK
- Scikit-learn
- Naive Bayes
Word2Vec Explorer
intermediateTrain or load a Word2Vec model to visualize and explore word similarities.
- Gensim
- Matplotlib
- PCA
Perceptrons, activation functions, forward/backward propagation
Loss functions, optimizers (SGD, Adam, RMSprop), learning rate
Feedforward networks, multilayer perceptrons, backpropagation
Conv layers, pooling, image classification, transfer learning
RNN, LSTM, GRU, sequence modeling, time series
PyTorch, TensorFlow/Keras, model building & training
Object detection (YOLO, R-CNN), image segmentation, GANs
Dropout, batch normalization, data augmentation
Pre-trained models, fine-tuning, domain adaptation
Customer Churn Prediction
intermediateTrain a Deep Neural Network (ANN) to predict if a customer will leave a subscription service.
- TensorFlow/Keras
- Pandas
Movie Review Sentiment
intermediateAnalyze the sentiment of movie reviews using a basic neural network architecture.
- PyTorch
- Embedding Layers
ML lifecycle, CI/CD for ML, model versioning
Self-attention, multi-head attention, scaled dot-product
Encoder-decoder, positional encoding, layer normalization
BERT, RoBERTa, ALBERT, DeBERTa
GPT, GPT-2, GPT-3, autoregressive language modeling
BPE, WordPiece, SentencePiece, Unigram
Implement transformer architecture from ground up
Build GPT from Scratch
advancedImplement the Transformer architecture and train a mini-GPT model token-by-token.
- PyTorch
- Python
- Math
AI vs ML vs DL vs GenAI, generative models overview
Pre-training, supervised fine-tuning, RLHF basics
Zero-shot, few-shot, chain-of-thought, prompt optimization
BLEU, ROUGE, perplexity, human evaluation
Bias, hallucinations, responsible AI, alignment
OpenAI, Anthropic, Cohere, Groq integration
AI Chatbot
beginnerBuild a chat interface that interacts with OpenAI or Anthropic models via API.
- Streamlit
- OpenAI API
Chains, agents, memory, callbacks
OpenAI, Ollama, Hugging Face, Groq in LangChain
Retrieval pipeline, embeddings, similarity search
FAISS, ChromaDB, Pinecone, Weaviate, Qdrant
sentence-transformers, OpenAI embeddings, custom embeddings
Multi-query retrieval, re-ranking, hybrid search
PDF parsing, text chunking, metadata extraction
Chat with Knowledge Base
intermediateCreate a RAG system that allows users to upload a PDF and ask questions about it.
- LangChain
- ChromaDB
- OpenAI
Document Summarizer
intermediateA tool that takes long text/documents and generates concise summaries.
- LangChain
- Hugging Face
Agent loops, planning, tool use, memory
Reasoning + Acting, chain-of-thought for agents
Function calling, external APIs, code execution
Agent collaboration, communication protocols
State machines, graph-based agent workflows
Managing multiple agents, workflow optimization
Autonomous Research Agent
advancedBuild an agent that searches the web, gathers information, and summarizes findings.
- LangGraph
- Tavily API
Automated Code Reviewer
advancedAn AI agent that reviews code snippets for bugs and suggests improvements.
- LangChain
- GitHub API
Customer Service Bot
advancedAn intelligent agent capable of routing queries and using tools to check order status.
- CrewAI
- Python
Why evals come first, logging and reading traces, error analysis to find failure modes, binary pass/fail criteria, and turning each failure into a regression test.
Direct scoring vs pairwise comparison, writing judge prompts and rubrics, position, verbosity and self-preference bias, and validating judges against human labels.
Retrieval metrics (hit rate, MRR, context precision/recall), faithfulness and answer relevance, tool-call and trajectory evaluation for agents, and offline vs online evaluation.
Traces and spans for LLM calls, retrieval and tool steps; logging prompts, outputs, tokens, latency and cost; and OpenTelemetry-based instrumentation.
- DOCLangfuse - LLM Observability & Application Tracing (docs)· free (open source, self-hostable; cloud free tier)
- DOCArize - Phoenix Documentation· free (open source)
- ARTOpenTelemetry - An Introduction to Observability for LLM-based applications using OpenTelemetry
- CRSHugging Face Agents Course - Bonus Unit: AI Agent Observability & Evaluation
Input and output validation, moderation APIs, safety classifiers like Llama Guard, programmable rails, running checks in parallel to limit latency, and failing safely.
Direct vs indirect prompt injection, jailbreak techniques, data exfiltration through tools, the lethal trifecta for agents, and why delimiters and "please ignore" instructions don't fix it.
Full fine-tuning vs transfer learning
Low-Rank Adaptation, quantized LoRA for efficient tuning
Prefix tuning, P-tuning, AdaLoRA, IA³
INT8, INT4, GPTQ, AWQ, GGML/GGUF
Pruning, knowledge distillation, QAT
- CRSMIT HAN Lab (Song Han) - MIT 6.5940 TinyML and Efficient Deep Learning Computing (Fall 2024)
- VIDUmar Jamil - Quantization explained with PyTorch (PTQ and QAT)
- ARTLilian Weng - Large Transformer Model Inference Optimization (distillation, pruning, quantization, sparsity)
- ARTNVIDIA - LLM Model Pruning and Knowledge Distillation with NeMo
Dataset preparation, instruction formatting
Reinforcement learning from human feedback overview
Fine-tune TinyLlama
advancedFine-tune the TinyLlama model on a custom instruction dataset.
- Unsloth
- PyTorch
- Hugging Face
Fine-tune Phi Model
advancedAdapt the Microsoft Phi SLM (Small Language Model) for a specialized task.
- PEFT
- LoRA
- BitsAndBytes
Protocol for connecting AI to external data and tools
Building MCP servers, exposing tools and data
Integrating MCP with AI assistants and applications
Weather MCP Server
advancedBuild an MCP server that connects to a weather API and exposes it to the Claude Desktop app.
- Python/TypeScript
- MCP SDK
- Weather API