A complete, hands-on guide to becoming an AI Engineer.
This repository is designed to help you learn AI from first principles, build real neural networks, and understand modern LLM systems end-to-end. You'll progress through math, PyTorch, deep learning, transformers, RAG, and OCR — with clean, intuitive Jupyter notebooks guiding you at every step.
Whether you're a beginner or an engineer levelling up, this repo gives you the clarity, structure, and intuition needed to build real AI systems.
If you learn something useful, a star is appreciated.
- Math functions, derivatives, vectors, and gradients
- Matrix operations and linear algebra
- Probability and statistics
- Creating and manipulating tensors
- Matrix multiplication, transposing, and reshaping
- Indexing, slicing, and concatenating tensors
- Special tensor creation functions
- Building neurons, layers, and networks from scratch
- Normalization techniques (RMSNorm)
- Activation functions
- Optimizers (Adam, Muon) and learning rate decay
- Attention and self-attention mechanisms
- Multi-head attention
- Decoder-only transformer architecture
- Building RAG pipelines end to end
- Indexing, retrieval, chunking strategies
- Integrations with embedding models and vector stores
- Cloud LLM support: Atlas Cloud (
deepseek-ai/DeepSeek-V3-0324by default), MiniMax (M3), OpenAI, or any OpenAI-compatible API for richer answer generation
- OCR pipeline and utilities
- Preprocessing images and extracting text
Beyond the numbered learning track, the repository also contains the following folders:
- LLM: LLM fundamentals, tokens, and token embeddings
- Fine-Tuning: LoRA, QLoRA, and Unsloth fine-tuning notebooks
- Basic ML Model Implementation: Classical supervised and unsupervised ML notebooks
- DL: Deep learning projects with production-style serving (FastAPI / Streamlit)
- Brazil-Ecommerce-Data-Analysis: Exploratory data analysis on the Olist dataset
- YoutubeMCP: A minimal MCP server for fetching YouTube transcripts
- Interview Prep, Books, Data, Start_here: Supporting resources and sample data
A detailed map of the repository:
ai-hands-on/
├── README.md # This file
├── LICENSE # MIT License
├── requirements.txt # Root Python dependencies
├── ANN_&_CNN.ipynb # ANN vs CNN on Fashion-MNIST (root copy)
├── ATLAS_CLOUD_PROVIDER_REVIEW.md # Notes on the Atlas Cloud provider integration in 5.RAG
├── test.txt # Sample network-intrusion CSV data
│
├── .github/
│ └── FUNDING.yml # Sponsorship config
│
├── Start_here/
│ └── learning_path.md # Recommended step-by-step learning path
│
├── 0.1-Do-you-know/ # Short "did you know" explainers
│ ├── ANN(Artificial_Neural_Network_).ipynb
│ └── Efficient_Iteration_In_Pandas.ipynb
│
├── 1.Math/ # Math fundamentals
│ ├── 1. Math Functions.ipynb
│ ├── 2. Partial Derivatives.ipynb
│ ├── 3. Vectors.ipynb
│ ├── 4. Gradient.ipynb
│ ├── 5. Metrics.ipynb
│ └── 6. Probability.ipynb
│
├── 2.Pytorch/ # PyTorch tensor basics
│ ├── 1. Tensor.ipynb
│ ├── 2. Matrix Multiplication.ipynb
│ ├── 3. Tensor Transposing.ipynb
│ ├── 4. Reshaping tensor.ipynb
│ ├── 5. Indexing & Slicing.ipynb
│ ├── 6. Tensor Concatenating.ipynb
│ └── 7. Linspace and Arrange Tensor.ipynb
│
├── 3.Neural-Network(NN)/ # Neural networks from scratch
│ ├── 1.Single Neuron.ipynb
│ ├── 2. Layer building.ipynb
│ └── 3. BackPropagation.md
│
├── 4.Transformer/ # Transformer building blocks
│ ├── 1. Attention Mechanism.ipynb
│ ├── 2. Self Attention.ipynb
│ ├── 3. Multi-head Attention.ipynb
│ └── 4. Decoder.ipynb
│
├── 5.RAG/ # Retrieval-Augmented Generation (cybersecurity use case)
│ ├── README.md
│ ├── .env.example # API key / provider configuration template
│ ├── requirements.txt
│ ├── data/processed_texts/ # Source documents (firewall, incident response, vulnerability scan)
│ ├── embeddings/faiss_index/ # Prebuilt FAISS index + metadata
│ ├── src/
│ │ ├── extract_text.py # Text extraction
│ │ ├── create_embeddings.py # Build embeddings and FAISS index
│ │ ├── retrieve_context.py # Semantic retrieval
│ │ ├── generate_answer.py # Answer generation
│ │ ├── llm_provider.py # Provider abstraction (Atlas Cloud, MiniMax, OpenAI, compatible APIs)
│ │ └── app.py # Application entry point
│ └── tests/ # test_integration.py, test_llm_provider.py
│
├── 6.OCR/ # Optical Character Recognition
│ ├── README.md
│ ├── main.py # API / app entry point
│ ├── utils.py # Image preprocessing and OCR helpers
│ ├── pytess_gemini.ipynb # Pytesseract + Gemini notebook
│ ├── render.yaml # Render deployment config
│ ├── requirements.txt
│ └── sample_data/ # Sample images (incl. a rotated one)
│
├── LLM/ # LLM fundamentals
│ ├── 1_LLM_Fundamentals_.ipynb
│ └── 2_Tokens_and_Token_Embeddings.ipynb
│
├── Fine-Tuning/ # Parameter-efficient fine-tuning
│ ├── Fine_Tuning_a_JSON_File_Using_Unsloth_&_LoRA.ipynb
│ ├── Lora-Implementation.ipynb
│ └── Qlora-Implementation.ipynb
│
├── Basic ML Model Implementation/ # Classical ML notebooks
│ ├── 1_Base_ML_Model_Linear_&_Ridge_Regression.ipynb
│ ├── LM_Implementation.ipynb
│ ├── Logistic_Regression_Model_Implementation.ipynb
│ ├── Decision_Tree_Model_Implementation.ipynb
│ ├── Naïve_Bayes_Classification_Implementation_.ipynb
│ ├── Support_Vector_Machine_[SVM]_Classification_Implementation_.ipynb
│ ├── Classification_Models_Implementation[Supervised].ipynb
│ └── Unsupervised_Learning_Models_Implementation.ipynb
│
├── DL/ # Deep learning projects (notebook -> production)
│ ├── ANN-CNN Prod/ # Fashion-MNIST ANN vs CNN served with FastAPI
│ │ ├── train.py, config.py
│ │ ├── app/ # FastAPI app, preprocessing, static upload page
│ │ └── models/ # Trained .keras models + metrics.json
│ ├── RNN-LSTM Prod/ # NumPy RNN vs LSTM from scratch + FastAPI + web UI
│ │ ├── seqnet/ # Models, optimizers, training, persistence, API
│ │ ├── web/ # Frontend (HTML/CSS/JS)
│ │ ├── artifacts/ # Trained weights
│ │ ├── tests/ # Gradient checks, API, notebook sync tests
│ │ └── Dockerfile, Makefile, start.sh
│ └── RNN-LSTM-GRU/ # HAR sensor data: RNN vs LSTM vs GRU
│ ├── backend/ # Config, data, model, training, FastAPI
│ ├── frontend/app.py # Streamlit UI
│ └── artifacts/ # Trained models, scaler, encoder, results
│
├── Brazil-Ecommerce-Data-Analysis/ # Olist e-commerce EDA
│ ├── README.md
│ └── customer / geolocation / order_items / order_payments /
│ order_reviews / orders / products / sellers _analysis.ipynb
│
├── YoutubeMCP/ # MCP server for YouTube transcripts
│ ├── README.md, Project.md
│ ├── main.py, server.py, test.py
│ ├── pyproject.toml
│ └── src/ # service.py, utils.py
│
├── Data/ # Sample datasets
│ ├── json_extraction_dataset_500.json # Used for JSON-extraction fine-tuning
│ └── test.txt
│
├── Interview Prep/
│ └── AI Engineering Topics To Prepare Must.html
│
└── Books/ # Reference PDFs (SQL, AI Engineering, Deep Learning, ML)
| Folder | Type | What you will find |
|---|---|---|
Start_here/ |
Guide | Learning path for going through the repo in order |
0.1-Do-you-know/ |
Notebooks | Quick explainers on ANNs and efficient Pandas iteration |
1.Math/ |
Notebooks | Functions, partial derivatives, vectors, gradients, metrics, probability |
2.Pytorch/ |
Notebooks | Tensor creation, matmul, transpose, reshape, indexing, concatenation, linspace/arange |
3.Neural-Network(NN)/ |
Notebooks + Notes | Single neuron, layer building, backpropagation notes |
4.Transformer/ |
Notebooks | Attention, self-attention, multi-head attention, decoder |
5.RAG/ |
Project | FAISS-based RAG pipeline with pluggable LLM providers and tests |
6.OCR/ |
Project | OCR service with preprocessing utilities and Render deployment |
LLM/ |
Notebooks | LLM fundamentals, tokens, and embeddings |
Fine-Tuning/ |
Notebooks | LoRA, QLoRA, and Unsloth fine-tuning |
Basic ML Model Implementation/ |
Notebooks | Regression, logistic regression, trees, Naive Bayes, SVM, unsupervised learning |
DL/ |
Projects | ANN/CNN, RNN/LSTM, and RNN/LSTM/GRU with APIs and UIs |
Brazil-Ecommerce-Data-Analysis/ |
Notebooks | EDA on customers, geolocation, orders, payments, reviews, products, sellers |
YoutubeMCP/ |
Project | MCP server that fetches YouTube transcripts |
Data/ |
Data | Sample datasets |
Interview Prep/ |
Resource | AI engineering interview topics checklist |
Books/ |
Resource | Reference PDFs |
Recommended reading to deepen your understanding (not included):
AI Engineeringby Chip HuyenHands-On Machine Learning with Scikit-Learn, Keras, and TensorFlowby Aurélien GéronDeep Learningby Ian Goodfellow, Yoshua Bengio, and Aaron CourvilleThe Elements of Statistical Learningby Trevor Hastie, Robert Tibshirani, and Jerome FriedmanNeural Networks and Deep Learningby Michael NielsenSQL Cookbookby Anthony Molinaro
For more books in AI/ML, I have created another repo for this Check Here. I will be adding lot more in coming days/months. If you are interested to read book, go check this repo out.
For a recommended step-by-step progression through the materials, see the Learning Path:
Start_here/learning_path.md
Install dependencies with:
pip install -r requirements.txtSome subfolders (for example 5.RAG/ and 6.OCR/) include their own requirements.txt with additional dependencies.
Recommended workflow:
-
Open Jupyter in the project root:
jupyter lab # or jupyter notebook -
Work through notebooks in order:
1.Math/2.PyTorch/3.Neural-Network(NN)/4.Transformer/
-
Folder to run separately:
5.RAG/6.OCR/DL/ANN-CNN Prod/,DL/RNN-LSTM Prod/,DL/RNN-LSTM-GRU/(each has its own README and requirements)YoutubeMCP/(requires Python 3.12+ anduv)
-
Resources
-
Basic ML Model Implementation (Supervised + Un-supervised + RL)
1.Linear Regression2.Logistic Regression3.Decision Tree Model4.Naive Bayes Classification
| Tool | Category | Link |
|---|---|---|
| Scikit-learn | Traditional ML | https://scikit-learn.org/stable/ |
| XGBoost | Gradient Boosting | https://xgboost.ai/ |
| LightGBM | Gradient Boosting | https://lightgbm.readthedocs.io/en/stable/ |
| CatBoost | Gradient Boosting | https://catboost.ai/ |
Contributions are welcome!
Please ensure:
- Notebooks are clean (Restart & Run All before committing)
- Existing structure & naming conventions are followed
- PRs are focused, readable, and documented
- In folders like RAG and OCR, please maintain the cleaned structure part
- If you want to add something new folders, make it proper structure way.
- This project is licensed under the MIT License. See
LICENSEfor details.
