Hands-On AI Engineering: A Collection of Practical AI Projects in Python

A Repository of Runnable AI Projects, Not a Textbook

Most AI learning resources take one of two forms: a textbook that explains concepts without runnable code, or a framework tutorial that teaches one provider's specific API. Hands-On AI Engineering takes a different approach: it is a collection of self-contained Python projects that each solve a specific problem and can be cloned, configured, and run.

The README describes the projects as production-ready and built to be adapted for real-world use. Each project includes complete code, setup instructions, and documentation. The target reader is a developer who wants to learn by running and modifying working implementations, not by reading theory.

The collection spans multiple modalities and multiple providers. Projects use OpenAI, Anthropic, Google, and open-source models, often through different frameworks in the same category. This means you can compare a LangGraph-based RAG system against an AG2-based multi-agent setup within the same repository by looking at different directories.

Repository Layout: Seven Top-Level Categories

The top-level directories are: ai_agents/, OCR/, rag_apps/, fine_tuning/, multimodal/, audio/, and assets/.

ai_agents/ is by far the largest category. The README lists over 20 distinct agent projects there, ranging from a multi-agent research assistant to a hotel finder, a brand monitor, a GitHub PR review agent, and a debate agent.

OCR/ covers optical character recognition systems.

rag_apps/ covers retrieval-augmented generation pipelines.

fine_tuning/ covers model fine-tuning projects.

multimodal/ covers projects that combine text with images, audio, or other modalities.

audio/ covers audio-based AI projects.

assets/ holds shared images and media used by the repository documentation.

Each project lives in its own subdirectory with its own code and README. There is no shared virtual environment or requirements file at the repository root; each project manages its own dependencies.

Cloning the Repository and Navigating to a Project

The README does not provide a single install command that sets up all projects. The intended workflow is to clone the repository and then navigate into the specific project you want to run:

git clone https://github.com/Sumanth077/Hands-On-AI-Engineering.git

From there, navigate into a specific project directory, such as:

cd ai_agents/research_assistant_with_memory

Each project subdirectory contains its own README with setup instructions, including which dependencies to install and which environment variables to configure. The projects use different model providers and frameworks, so setup steps vary by project.

The repository's homepage links to a newsletter at aiengineering.beehiiv.com, which appears to be the companion publication where new projects are announced and explained.

What the AI Agents Category Contains

The ai_agents/ directory holds the largest group of projects in the repository. The README lists them explicitly. A representative selection:

The Multi-Agent Research Assistant with Memory uses Ollama for local inference and BGE embeddings with an Actian VectorAI DB for the memory layer. It retrieves cited answers from PDFs, papers, manuals, and transcripts, and persists findings across sessions.

The Daily AI News Digest uses MiniMax M2.7 to score articles from 92 technology blogs curated by Andrej Karpathy, surfacing the three most significant stories from the past 24 hours and delivering them to Telegram.

The Self-Reflective Agentic RAG project uses LangGraph. It grades retrieved context quality, rewrites the query if the context does not pass validation, and only generates an answer once the context is deemed sufficient.

Eagle Eye is a GitHub PR review agent that fetches diffs via GitHub MCP, performs structured code review with severity ratings, and posts feedback after user approval. It is built with OpenClaw and Telegram.

The CartMate customer support agent uses Mem0 and Mistral Small 4 to remember customers across sessions and continue conversations where they left off.

The projects cover a range of patterns: single agents, multi-agent pipelines, memory-augmented systems, and tool-using agents. Each uses its own model provider and framework, reflecting the breadth rather than depth approach of the collection.

RAG Apps and OCR: The Other Major Categories

The rag_apps/ directory covers retrieval-augmented generation. The Self-Reflective Agentic RAG project (listed in ai_agents/ but representative of this category's pattern) shows a complete pipeline: retrieval, context grading, query rewriting on failure, and answer generation. The README does not enumerate all rag_apps/ projects individually in the main README text.

The OCR/ directory covers optical character recognition projects. The README lists OCR as one of the top-level project categories and it appears in the directory listing, but the main README does not enumerate the individual OCR projects in the way that ai_agents/ projects are listed.

The fine_tuning/ and multimodal/ directories follow the same structure: each contains self-contained projects with their own setup. The multimodal category covers projects that work with multiple input types beyond text alone.

Where the Collection Falls Short

Three limitations affect how you should use this repository.

First, there is no single entry point. A developer who wants to explore the full collection must navigate into individual project directories, read separate READMEs, and configure each project's environment independently. There is no shared scaffold or common utilities that reduce setup overhead across projects.

Second, the projects depend on external services and API keys. Most projects require at least one paid API key (OpenAI, Anthropic, Google, or a commercial model provider) or a locally running model server (Ollama). A developer without any of these set up will not be able to run most projects without additional configuration.

Third, there are no GitHub release tags. The repository is a living collection updated regularly, but code you clone today may differ from code you cloned last month. Projects may be updated, removed, or have their dependencies changed between clones. The README does not document a changelog or version history for individual projects.

Hands-On AI Engineering vs LangChain Cookbook: Breadth Against Framework Depth

LangChain Cookbook is a repository of example notebooks and scripts built around the LangChain framework. Its examples teach LangChain-specific patterns: chains, agents, memory, retrievers, and tools as LangChain implements them.

Hands-On AI Engineering is framework-agnostic. Projects use LangChain, LangGraph, AG2, smolagents, Agno, and direct API calls to different providers. The variety means you see how different frameworks solve the same class of problem, but it also means there is no consistent interface or pattern to learn across projects.

For a developer who has chosen LangChain and wants to go deep on it, the Cookbook is more focused. For a developer who wants to survey the current landscape of agent frameworks and model providers without committing to one, Hands-On AI Engineering shows more options in a single repository.

Editorial conclusion

Hands-On AI Engineering is a good starting point for developers who want working, runnable examples of AI agents, RAG systems, and OCR pipelines without committing to a single provider or framework. The breadth of providers (OpenAI, Anthropic, Google, open-source models via Ollama) means you can evaluate different approaches side by side. The repository has no stable release tags, so the code is a living snapshot that may change between clones. Verify that the specific project you want to run has its own setup instructions, since the projects use different dependencies and model providers and there is no single shared environment.

Frequently asked questions

Official sources

1. Issues
2. Project website
3. README
4. Sumanth077/Hands-On-AI-Engineering on GitHub