NarimanN2/ollama-playground

What it solves

This repository fills the gap between learning about Ollama's open-source models and actually building with them. It provides a set of ready-made example projects that show how to use these models for real tasks like chatting with PDFs, building AI agents, and creating voice or vision applications. Instead of just reading documentation, you can see working code and watch tutorial videos.

How it works

The repo is organized as a set of independent projects, each with its own folder and README. Each project focuses on a specific use case and uses Ollama models as the core engine. The projects are grouped into categories:

• Retrieval-Augmented Generation (RAG) – including chat with PDFs, hybrid retrieval, multimodal RAG, and voice RAG.
• Agent Tooling & Protocols – showing how to build agents with memory, MCP-based and ACP-based agents, and secure MCP servers.
• Practical Agents – like web scraping, research assistant, and text-to-SQL agents.
• Multi-Agent Systems – with supervisor and swarm architectures.
• Voice – including an AI podcast generator and a local voice assistant.
• Vision – video summarization, OCR, emotion detection, object detection, and an image search engine. Each project comes with code that you can run locally, and each is accompanied by a YouTube video that explains how it was built.

Who it's for

This is for developers, hobbyists, and students who want to learn how to build applications with locally running open-source LLMs. It's also useful for anyone interested in practical implementations of RAG, agents, voice, and vision without having to start from scratch. The video tutorials make it easy to follow along, even if you're new to these concepts.

Highlights

• Covers a wide range of use cases: RAG, agents, voice, vision, and multi-agent systems
• Each project includes a dedicated README and links to YouTube tutorials
• Projects demonstrate both simple and advanced architectures (supervisor, swarm, MCP, ACP)
• Includes multimodal and voice interfaces, showing how to combine multiple input/output types
• Everything is built on Ollama's open-source models, so you can run it locally without cloud dependencies