AI/ML Learning Checklist βœ…ΒΆ

START HERE β€” Essential Reading (Do This First)ΒΆ

  • Done MASTER_STUDY_GUIDE.md β€” Phase-by-phase learning notes, text explanations, and study schedule. Start here.

  • Done CAREER_ROADMAP.md β€” Target roles, interview prep, portfolio strategy, and 30-60-90 day plan.

  • Done REFERENCES.md β€” All videos, GitHub repos, courses, papers, and tools organized by phase.

  • Done INTERVIEW_PREP.md β€” 15 ML Q&A with answers, specific coding problems, 3 system design walkthroughs.

πŸ“‹ How to Use This ChecklistΒΆ

Phase 0: Glossary & FoundationsΒΆ

Phase 1: Python & Machine Learning (278 notebooks)ΒΆ

Python FoundationsΒΆ

Scikit-learn Mastery (278 notebooks)ΒΆ

  • Done 02-data-science/5-scikit-learn/ - 278 example notebooks

  • Practice linear regression examples

  • Practice classification examples (SVM, Random Forest, etc.)

  • Practice clustering examples (K-means, DBSCAN, etc.)

  • Practice dimensionality reduction (PCA, t-SNE, etc.)

Microsoft LabsΒΆ

ProjectsΒΆ

  • Build a classification model (e.g., Iris dataset)

  • Build a regression model (e.g., housing prices)

  • Build a clustering project (e.g., customer segmentation)

Phase 2: Mathematics for MLΒΆ

Linear AlgebraΒΆ

Calculus & DerivativesΒΆ

Probability & StatisticsΒΆ

Optimization & Gradient DescentΒΆ

ISLP Book (13 notebooks)ΒΆ

CS229 Course NotebooksΒΆ

Video CoursesΒΆ

Phase 3: Tokenization (8 notebooks)ΒΆ

Tokenizer NotebooksΒΆ

DocumentationΒΆ

Phase 4: Embeddings (10 notebooks)ΒΆ

Core Embedding NotebooksΒΆ

Advanced TopicsΒΆ

DocumentationΒΆ

Phase 5: Neural NetworksΒΆ

Core NotebooksΒΆ

Math FoundationΒΆ

Microsoft LabsΒΆ

Video CoursesΒΆ

  • Done 3Blue1Brown - Neural Networks series (see REFERENCES.md)

  • Done Andrej Karpathy - Neural Networks: Zero to Hero (see REFERENCES.md)

  • Done MIT 6.S191 - Intro to Deep Learning (see REFERENCES.md)

ProjectsΒΆ

  • Build a neural network from scratch using only NumPy (no PyTorch)

  • Implement backpropagation by hand for a 2-layer network

  • Train a transformer on a character-level language modeling task

Phase 6: Vector Databases (7 notebooks)ΒΆ

Core NotebooksΒΆ

DocumentationΒΆ

ProjectsΒΆ

  • Build a semantic search engine with Chroma

  • Deploy a vector database locally

  • Benchmark different vector databases

Phase 7: RAG SystemsΒΆ

Core RAG NotebooksΒΆ

DocumentationΒΆ

Microsoft LabsΒΆ

ProjectsΒΆ

  • Build a basic RAG chatbot over a PDF

  • Build a document Q&A system with hybrid search + reranking

  • Implement RAG with RAGAS evaluation metrics

  • Build a fully local RAG (Ollama + ChromaDB, no cloud APIs)

Phase 8: MLOpsΒΆ

Core NotebooksΒΆ

DocumentationΒΆ

Video CoursesΒΆ

  • Watch Made With ML - MLOps (free at madewithml.com)

  • Watch Weights & Biases tutorials for experiment tracking

ProjectsΒΆ

  • Deploy a classifier with FastAPI + Docker + GitHub Actions CI/CD

  • Set up MLflow experiment tracking for any training run

  • Deploy a local LLM server with vLLM (OpenAI-compatible endpoint)

Phase 9: SpecializationsΒΆ

Path A: Computer Vision (best for CV Engineer / Multimodal AI roles)ΒΆ

Path B: NLP (best for NLP Engineer / Text AI roles)ΒΆ

Path C: AI Agents Specialization (best for AI Engineer roles β€” pairs with Phase 14)ΒΆ

Microsoft LabsΒΆ

Phase 10: Prompt Engineering (6 notebooks) πŸ”₯ΒΆ

Core NotebooksΒΆ

DocumentationΒΆ

Video CoursesΒΆ

ProjectsΒΆ

  • Build a ReAct agent with tools

  • Create a prompt optimization system

  • Build a multi-step reasoning chain

Phase 11: LLM Fine-tuning (12 notebooks) πŸ”₯ΒΆ

Core NotebooksΒΆ

DocumentationΒΆ

Video CoursesΒΆ

  • DeepLearning.AI - Finetuning Large Language Models

  • Hugging Face - Transformers Course

ProjectsΒΆ

  • Fine-tune a model with LoRA on a custom dataset

  • Apply DPO alignment to improve helpfulness

  • Deploy a fine-tuned model with vLLM

  • Compare QLoRA vs DoRA quality vs memory tradeoffs

Phase 12: Multimodal AI πŸ”₯ΒΆ

Vision-Language ModelsΒΆ

Image GenerationΒΆ

DocumentationΒΆ

ProjectsΒΆ

  • Build an image search engine: embed images with CLIP, query by text

  • Create a text-to-image generator with custom prompts and negative prompts

  • Build a multimodal chatbot that can describe and discuss images (use GPT-5.4, Gemini 3.1, or Qwen2.5-VL locally)

Phase 13: Local LLMs (6 notebooks) πŸ”₯ΒΆ

Core NotebooksΒΆ

DocumentationΒΆ

Video CoursesΒΆ

  • Matthew Berman - Local LLMs

ProjectsΒΆ

  • Set up Ollama and run Qwen 3 / Llama 4 locally

  • Build a fully local private RAG system over your documents

  • Deploy a local LLM server with OpenAI-compatible API

Phase 14: AI Agents (9 notebooks) πŸ”₯ HOT TOPICΒΆ

Core NotebooksΒΆ

DocumentationΒΆ

Pre/Post AssessmentΒΆ

ProjectsΒΆ

  • Build a research agent with web search + calculator + code execution tools

  • Build a multi-agent pipeline: planner β†’ executor β†’ reviewer

  • Build an MCP-powered agent connected to your own data sources

  • Create a LangGraph stateful agent with persistent memory

Phase 15: Real-Time Streaming AI (4 notebooks) πŸ”₯ΒΆ

Core NotebooksΒΆ

DocumentationΒΆ

ProjectsΒΆ

  • Build a ChatGPT-style streaming chat interface with FastAPI + SSE

  • Create a real-time RAG system that streams sources then answers

  • Deploy a production streaming server with monitoring dashboards

Supplementary Phases (Do These Alongside or After Core Track)ΒΆ

Model Evaluation β€” 16-model-evaluation/ΒΆ

Debugging & Troubleshooting β€” 17-debugging-troubleshooting/ΒΆ

Low-Code AI Tools β€” 18-low-code-ai-tools/ΒΆ

AI Safety & Red Teaming β€” 19-ai-safety-redteaming/ΒΆ

πŸ§ͺ Hands-On LabsΒΆ

Microsoft Labs (100+ lessons)ΒΆ

Video Courses (50+ channels)ΒΆ

Cloud PlatformsΒΆ

πŸŽ“ MDTP Framework - Comprehensive Knowledge MapΒΆ

M - ModelsΒΆ

M0. Statistical Thinking & IntuitionΒΆ

  • Probability, random variables, distributions

  • Model vs. reality, digital twins

  • Structure and causality

  • Inference vs generalization (Breiman’s β€œTwo Cultures”)

  • Observation, experiment and experiment design

  • Treatment and treatment effects

  • Measurement errors and missing data

M1. Statistical InferenceΒΆ

  • Data generating process, sample vs population

  • Sampling techniques and inference

  • Frequentist and Bayesian views

  • Measuring model quality and performance

  • Model lifecycle, data and model drift

M2. EconometricsΒΆ

  • Cross-section, time series, panel and spatial data

  • Linear regression and OLS

  • Violation of OLS assumptions (Peter Kennedy’s 10 Commandments)

  • Estimation techniques: OLS, GMM, maximum likelihood, Bayesian

M4. Classic Machine LearningΒΆ

  • Classification, regression, clustering

  • Dimensionality reduction, decision trees

  • Support vector machines and discriminant analysis

  • Model selection and performance evaluation

M5. Neural Networks & Deep LearningΒΆ

  • Simple perceptron and NN construction

  • Gradient descent, backpropagation, regularization

  • Traditional architectures (feed-forward, CNN, RNN, GAN, transformers)

  • Embeddings and vector representations (word2vec, sentence-transformers)

  • Tokenization (BPE, WordPiece, SentencePiece)

  • Attention mechanism (self-attention, multi-head attention)

  • Mixture of Experts (MoE) architecture

M6. NLP, CV and Advanced SubfieldsΒΆ

  • Classic NLP (Jurafsky and Martin)

  • Large Language Models (GPT-4, Claude, LLaMA, Gemini)

  • Transformer architecture, prompt engineering

  • RAGs and fine-tuning

  • Instruction tuning and alignment (RLHF, DPO, constitutional AI)

  • In-context learning (few-shot, zero-shot, chain-of-thought)

  • Model compression (quantization: 4-bit, 8-bit; distillation; LoRA/QLoRA)

  • Hallucination detection and mitigation

  • Computer vision fundamentals

  • Multimodal models (CLIP, GPT-4V, vision-language)

  • Diffusion models (Stable Diffusion, DALL-E, Midjourney)

M9. Modern AI Architectures (2020-2025)ΒΆ

  • Transformer variants (BERT, GPT, T5)

  • Vision Transformers (ViT) for image understanding

  • State Space Models (Mamba, structured state spaces)

  • Retrieval-augmented architectures (RAG systems)

  • Sparse models and efficient transformers (Sparse Transformers, Longformer)

D - DataΒΆ

D1. Data SourcesΒΆ

  • Data collection, observation vs experiment

  • Physical sensors

  • Proprietary vs open data, official statistics

  • Data distribution and providers

  • Data protection and privacy

D2. Data AnalysisΒΆ

  • EDA and descriptive statistics

  • Graphs, visualizations, dashboards

  • Analysis as a DAG

  • Reproducible research

D3. Data EngineeringΒΆ

  • Structured data, serialization formats (CSV, JSON, XML)

  • SQL for tabular data

  • Dataframes (pandas, polars)

  • Processing large datasets (MapReduce, Hadoop, Spark)

  • Everything as a vector: text, images, sound, video

  • Data ingestion, transform, storage, retrieval

  • Vector databases (Pinecone, Weaviate, ChromaDB, pgvector)

  • Embedding storage and retrieval (semantic search, HNSW, FAISS)

  • Real-time data streaming (Kafka, Flink)

  • Data versioning (DVC, LakeFS)

D4. Pipelines & OrchestrationΒΆ

  • Modelling pipelines (Airflow)

  • Model delivery (FastAPI)

  • MLOps platforms (MLflow, Weights & Biases, Kubeflow, Neptune)

  • Feature stores (Feast, Tecton)

  • A/B testing and experimentation frameworks

  • Model versioning and registry

D5. Data ManagementΒΆ

  • Data quality assurance

  • Corporate data governance (DMBOK)

T - Tools (Code & Infrastructure)ΒΆ

T0. Writing CodeΒΆ

  • Linux and command line (local and remote)

  • Software development practices (version control, unit testing, APIs)

  • DevOps, product thinking and metrics

  • Python, R, Julia ecosystems

T1. Software ToolsΒΆ

  • Statistical packages and documentation

  • R for statistics, scikit-learn for ML

  • PyTorch, TensorFlow, Keras for neural networks

  • LLM frameworks (LangChain, LlamaIndex, Haystack)

  • Hugging Face ecosystem (Transformers, Datasets, Accelerate, PEFT)

  • Vector database libraries (FAISS, Annoy, hnswlib)

  • Modern ML frameworks (JAX, XGBoost, LightGBM, CatBoost)

T2. Cloud & Computing InfrastructureΒΆ

  • Cloud services and infrastructure provisioning

  • GPU computing (CUDA, distributed training, multi-GPU)

  • Container orchestration (Docker, Kubernetes for ML)

  • Serverless ML (AWS Lambda, Modal, Banana, Replicate)

  • Edge deployment (TensorFlow Lite, ONNX Runtime)

  • LLM hosting (vLLM, TGI, Ollama for local models)

T2.1. Understanding the Cloud (Business Perspective)ΒΆ

  • Client-server foundation and architecture

  • Virtualization and resource efficiency

  • Storage solutions (S3, block, object storage)

  • Containerization and orchestration (Kubernetes)

  • PaaS and serverless (Lambda)

  • Networking: zones, CDNs, load balancing

  • On-premise vs cloud tradeoffs

  • Economics of cloud computing

  • Cloud providers: AWS, GCP, Azure

  • Data center energy efficiency

T2.2. Databases & StorageΒΆ

  • HDD vs SSD vs Cloud (S3) tradeoffs

  • File systems (HDFS)

  • Relational databases (PostgreSQL, MySQL)

  • NoSQL types: Key-Value, Document, Column, Graph

  • Vector databases (Pinecone, Weaviate, ChromaDB)

  • Time series databases (InfluxDB, TimescaleDB)

  • Large data processing (MapReduce, Hadoop, Spark)

  • Search databases (ElasticSearch, Solr)

  • Data warehouses (Snowflake, Databricks, BigQuery)

  • Database theory: ACID, CAP theorem, BASE

  • OLAP vs OLTP

T2.3. Data Engineering ToolsΒΆ

  • Data engineering lifecycle: Generation β†’ Storage β†’ Ingestion β†’ Transformation β†’ Serving

  • Workflow orchestration (Airflow, Prefect, Dagster)

  • MLFlow for ML lifecycle management

  • Modern data stack components

  • Ingestion: Fivetran, Airbyte, Stitch

  • Transform: dbt, Spark

  • BI tools: Looker, Tableau, PowerBI

T3. Development Tools & EnvironmentΒΆ

  • Jupyter ecosystem (JupyterLab, notebooks, extensions)

  • IDEs and AI coding assistants (VS Code, Cursor, GitHub Copilot)

  • Experiment tracking (Weights & Biases, Neptune, Comet)

  • Data labeling tools (Label Studio, Prodigy, Scale AI)

P - ProductisationΒΆ

P1. Productisation & Business ValueΒΆ

  • Risk, learning and experimentation in business context

  • Modelling hypothesis and expected outcomes

  • Data-model-action workflows

  • Team roles and hiring

  • Measuring business outcomes and ML impact

  • RAG systems architecture and implementation

  • Prompt engineering as systematic discipline

  • AI agent frameworks (AutoGPT, LangChain agents, function calling)

  • Model monitoring (drift detection, performance tracking)

  • Cost optimization (token usage, caching, batch processing)

P1.1. ML Project Lifecycle FrameworkΒΆ

  • A. Identify Business Case - Hypothesis and expected ROI

  • B. Create Adequate Model - Value chain representation

  • C. State Proposed Change - Why, what, how, success criteria

  • D. Prove with Experiments - Run controlled tests

  • E. Scale or Pivot - Based on results

  • F. Continuous Improvement - Feedback loops

P1.2. Data Team Roles & ResponsibilitiesΒΆ

  • Understand different role functions:

    • Full-Stack Data Scientist

    • Data Scientist/Modeler

    • Data Engineer/Architect

    • Machine Learning Engineer

    • Research Scientist

    • Business Analyst

    • Product Manager

P1.3. Production ChallengesΒΆ

  • Technical: Data quality, model drift, infrastructure failures

  • Organizational: Silos, unclear responsibilities

  • Business: Unclear success criteria, ROI validation

P2. Applications, Domains & CasesΒΆ

  • Recommender systems (RecSys)

  • Clinical trials

  • Quality control and industrial automation

  • Finance applications

  • Conversational AI and chatbots

  • Code generation (GitHub Copilot, code assistants)

  • Document intelligence (Q&A, summarization, extraction)

  • Search and information retrieval (semantic search, hybrid search)

P3. Society Impacts & RegulationΒΆ

  • Fairness, biases, equity, ethics

  • Grounds for AI regulation and policy

  • AI safety and alignment (value alignment, red teaming)

  • Prompt injection and security vulnerabilities

  • Environmental impact (carbon footprint of training)

  • AI watermarking and detection

  • Copyright and IP issues with generative AI

πŸ“š Documentation & ResourcesΒΆ

🎯 Final Projects (Build Your Portfolio)¢

Beginner ProjectsΒΆ

  • Iris classification (scikit-learn)

  • House price prediction (regression)

  • Customer segmentation (clustering)

Intermediate ProjectsΒΆ

  • Sentiment analysis (NLP)

  • Image classifier (CNN)

  • Recommendation system (collaborative filtering)

Advanced ProjectsΒΆ

  • RAG chatbot for your documents

  • Fine-tuned LLM for specific domain

  • Multimodal AI application (vision + language)

Production ProjectsΒΆ

  • Deploy ML model with FastAPI + Docker

  • Build CI/CD pipeline for ML

  • Create monitoring dashboard for ML models

⭐ Completion Milestones¢

  • Foundation Complete - Phases 0-2 done (Python, Math, ML basics)

  • Modern AI Stack - Phases 3-6 done (Tokenization, Embeddings, Neural Nets, Vector DBs)

  • Production Ready - Phases 7-9 done (RAG, MLOps, Specializations)

  • Cutting-Edge AI - Phases 10-13 done (Prompt Engineering, Fine-tuning, Multimodal, Local LLMs)

  • Advanced Research - Phases 24-25 done (Advanced Math, Deep Learning Research)

  • Portfolio Built - 5+ projects deployed and documented

  • Certified - At least 1 ML certification (AWS, Google, Azure, or DeepLearning.AI)

Phase 24: Advanced Deep Learning (RESEARCH LEVEL) πŸ”¬ΒΆ

Generative Models - AdvancedΒΆ

Advanced Topics (To Be Implemented)ΒΆ

  • Info-GAN and Conditional GANs

  • Bayesian GANs

  • Advanced VAE architectures

  • 3D Vision and NeRF

  • Vision Transformers (ViT)

  • Advanced Transformer architectures

Phase 25: Reinforcement Learning (7 notebooks)ΒΆ

Core NotebooksΒΆ

Advanced TopicsΒΆ

  • Policy gradients and actor-critic methods (planned)

  • Deep reinforcement learning (planned)

  • Multi-agent reinforcement learning (planned)

Phase 26: Time Series Analysis & Forecasting (7 notebooks)ΒΆ

Core NotebooksΒΆ

Advanced TopicsΒΆ

  • GARCH models for volatility (planned)

  • State space models (planned)

  • Bayesian time series (planned)

Phase 27: Causal Inference & Experimental Design (7 notebooks)ΒΆ

Core NotebooksΒΆ

Advanced TopicsΒΆ

  • Mediation analysis (planned)

  • Causal discovery (planned)

  • Difference-in-differences with multiple periods (planned)

πŸ† Completion MilestonesΒΆ

  • Foundation Complete - Phases 0-2 done (Python, Math, ML basics)

  • Modern AI Stack - Phases 3-6 done (Tokenization, Embeddings, Neural Nets, Vector DBs)

  • Production Ready - Phases 7-9 done (RAG, MLOps, Specializations)

  • Cutting-Edge AI - Phases 10-13 done (Prompt Engineering, Fine-tuning, Multimodal, Local LLMs)

  • Research Mastery - Phases 24-27 done (Advanced Deep Learning, Reinforcement Learning, Time Series, Causal Inference)

  • Portfolio Built - 5+ projects deployed and documented

  • Certified - At least 1 ML certification (AWS, Google, Azure, or DeepLearning.AI)

πŸŽ“ Next Steps After CompletionΒΆ

  • Contribute to open-source ML projects (start with docs/issues on popular repos)

  • Write blog posts about what you built (1 post per major project)

  • Mentor others in AI/ML

  • Build your own AI product

  • Apply for ML/AI jobs or research positions

  • Continue learning (AI never stops evolving!)

  • Read and implement papers from top conferences (NeurIPS, ICML, ICLR)

  • Contribute to ML research

Career Readiness MilestonesΒΆ

  • Resume Updated β€” AI/ML skills section added, projects listed

  • LinkedIn Updated β€” Headline reflects AI/ML focus, summary updated

  • GitHub Polished β€” 4+ repos pinned, each with README + demo

  • Project 1 Live β€” RAG chatbot deployed (HF Spaces / Render)

  • Project 2 Live β€” Fine-tuned model with evaluation results documented

  • Project 3 Live β€” MLOps pipeline or AI agent

  • Interview Prep β€” Practiced 10 core ML concept questions

  • Coding Practice β€” Solved 20+ pandas/sklearn coding problems

  • System Design β€” Practiced 3 ML system design scenarios

  • Networking β€” Connected with 10+ AI engineers on LinkedIn

  • Applications Sent β€” Applied to 20+ positions