# AI-Engineering-from-Scratch Dependencies: Python, TypeScript, Rust, and Julia Libraries > Discover the essential dependencies for AI Engineering from Scratch. Explore Python, TypeScript, Rust, and Julia libraries required for this comprehensive project. - Repository: [Rohit Ghumare/ai-engineering-from-scratch](https://github.com/rohitg00/ai-engineering-from-scratch) - Tags: getting-started - Published: 2026-06-06 --- **The ai-engineering-from-scratch project uses a multi-language dependency structure with Python packages centralized in [`requirements.txt`](https://github.com/rohitg00/ai-engineering-from-scratch/blob/main/requirements.txt), TypeScript packages declared in capstone-specific [`package.json`](https://github.com/rohitg00/ai-engineering-from-scratch/blob/main/package.json) files, and zero external dependencies for Rust and Julia lessons.** The rohitg00/ai-engineering-from-scratch curriculum is a comprehensive educational repository that teaches AI engineering through four programming languages. Understanding the complete ai-engineering-from-scratch dependencies is essential for running the hands-on lessons, which follow a deliberate progression from zero-dependency implementations to production-grade frameworks. ## Python Dependencies in requirements.txt The Python stack consolidates all dependencies in the root-level **[`requirements.txt`](https://github.com/rohitg00/ai-engineering-from-scratch/blob/main/requirements.txt)**. This manifest bundles production-grade machine learning frameworks alongside data science utilities. Core packages include: - **Deep learning**: `torch`, `torchvision`, `torchaudio`, `transformers`, `datasets`, `tokenizers`, `accelerate` - **Data science**: `numpy`, `pandas`, `scikit-learn`, `matplotlib` - **Multimedia processing**: `pillow`, `librosa`, `soundfile` - **LLM tooling**: `openai`, `anthropic`, `tiktoken` - **Development environment**: `jupyter` ## TypeScript Dependencies for Capstone Projects Unlike the centralized Python approach, TypeScript dependencies are decentralized across capstone lessons. Each project maintains its own manifest under **`phases/19-capstone-projects/*/code/ts/package.json`**. Common runtime dependencies include: - **`zod`** – Schema validation used across multiple capstones for type safety - **`hono`** – Web framework utilized in specific lessons such as the RAG over codebase implementation To install TypeScript dependencies for a specific lesson, navigate to the lesson directory and run: ```bash cd phases/19-capstone-projects/01-terminal-native-coding-agent/code/ts npm install npm run start ``` ## Rust and Julia: Standard Library Only The curriculum enforces a **standard-library-only** policy for Rust and Julia implementations. **Rust** lessons rely exclusively on the Rust standard library with **no external crates**. This restriction is formally defined in **[`AGENTS.md`](https://github.com/rohitg00/ai-engineering-from-scratch/blob/main/AGENTS.md)** under the Allowed-Dependencies policy. **Julia** lessons similarly prohibit third-party packages, utilizing only built-in modules such as `Random`, `Statistics`, `LinearAlgebra`, and `Printf`. ## How to Install the Dependencies ### Python Environment Setup Create a virtual environment and install the full Python stack: ```bash python3 -m venv .venv source .venv/bin/activate pip install -r requirements.txt ``` Alternatively, use the provided helper script: ```bash python scripts/install_skills.py ``` This script installs the Python dependencies and links the generated skills for the complete curriculum. ### TypeScript Environment Setup For each capstone requiring Node.js: ```bash cd phases/19-capstone-projects/[lesson-name]/code/ts npm i ``` ### Running Rust Lessons No installation required beyond the Rust compiler: ```bash rustc path/to/lesson/main.rs -O && ./main ``` ### Running Julia Lessons Execute directly with the Julia runtime using only standard library modules: ```bash julia path/to/lesson/main.jl ``` ## Dependency Policy and Governance The **[`AGENTS.md`](https://github.com/rohitg00/ai-engineering-from-scratch/blob/main/AGENTS.md)** file establishes the philosophical boundary between "Build-It" and "Use-It" phases. This policy mandates that early algorithm implementations use zero external libraries (enforced for Rust and Julia), while later phases demonstrate production equivalents using frameworks like PyTorch and Hono. This intentional layering ensures conceptual clarity before introducing abstraction complexity. ## Summary - **Python**: All dependencies live in root [`requirements.txt`](https://github.com/rohitg00/ai-engineering-from-scratch/blob/main/requirements.txt), covering PyTorch, Transformers, OpenAI, and data science stacks - **TypeScript**: Distributed across `phases/19-capstone-projects/*/code/ts/package.json`, primarily using `zod` and occasionally `hono` - **Rust**: Zero external crates; standard library only per [`AGENTS.md`](https://github.com/rohitg00/ai-engineering-from-scratch/blob/main/AGENTS.md) policy - **Julia**: No third-party packages; relies on `Random`, `Statistics`, `LinearAlgebra`, and `Printf` - **Automation**: Use [`scripts/install_skills.py`](https://github.com/rohitg00/ai-engineering-from-scratch/blob/main/scripts/install_skills.py) for one-command Python setup ## Frequently Asked Questions ### Where are the Python dependencies defined in ai-engineering-from-scratch? All Python dependencies are declared in the **[`requirements.txt`](https://github.com/rohitg00/ai-engineering-from-scratch/blob/main/requirements.txt)** file at the repository root. This single manifest contains the complete set of libraries needed for the Python-based lessons, including PyTorch, Transformers, and various data science tools. The file serves as the source of truth for the entire Python curriculum. ### Do the TypeScript capstone projects share a single package.json? No. Unlike the Python setup, each TypeScript capstone maintains its own **[`package.json`](https://github.com/rohitg00/ai-engineering-from-scratch/blob/main/package.json)** under `phases/19-capstone-projects/[specific-lesson]/code/ts/`. While many share common dependencies like `zod` for validation, each lesson directory must have its dependencies installed individually via `npm install`. ### Why don't the Rust lessons use external crates? The repository follows a **"stdlib-first"** philosophy enforced by the Allowed-Dependencies policy in [`AGENTS.md`](https://github.com/rohitg00/ai-engineering-from-scratch/blob/main/AGENTS.md). This design forces learners to understand core algorithms without abstraction layers before graduating to framework-based implementations in other languages. Consequently, all Rust lessons compile with only the standard library using commands like `rustc path/to/lesson/main.rs -O`. ### Is there a helper script to install all Python dependencies at once? Yes. The repository includes **[`scripts/install_skills.py`](https://github.com/rohitg00/ai-engineering-from-scratch/blob/main/scripts/install_skills.py)**, which automates the installation of all Python packages listed in [`requirements.txt`](https://github.com/rohitg00/ai-engineering-from-scratch/blob/main/requirements.txt) and links the generated skills. This script provides the fastest path to getting the complete Python environment running for the entire curriculum.