Building AI Applications With Haystack

This repository contains the notebooks and hands-on materials for the course "Building AI Applications With Haystack", provided by DeepLearning.AI in collaboration with Haystack.

The course teaches you how to use Haystack, an open-source framework, to build powerful retrieval-augmented generation (RAG) pipelines, intelligent agents, and custom AI applications.

---

📚 Course Topics

1️⃣ Haystack Building Blocks

In this lesson, you’ll explore the fundamental building blocks of Haystack:

• Components – the modular units of computation (retrievers, readers, generators, etc.).
• Running Components Individually – test each piece before combining.
• Combining Components – link them together to form data flows.
• Pipelines – construct workflows for different AI tasks.
• Document Search Pipeline – build your first simple search system.

This provides a solid foundation for understanding how everything in Haystack fits together.

---

2️⃣ Build Customized RAG

Here, you’ll go beyond the basics and create your own Retrieval-Augmented Generation (RAG) pipeline:

• Combine retrievers, readers, and LLMs into a RAG workflow.
• Customize the behavior of RAG (e.g., change how many documents to retrieve, tweak prompts).
• Understand how to adapt RAG to different tasks, such as answering factual questions or summarizing documents.

By the end, you’ll be comfortable designing tailored RAG pipelines for your own applications.

---

3️⃣ Custom Components – News Summarizer

This lab shows you how to extend Haystack by writing custom components:

• Build your own custom component in Python.
• Create a News Summarizer component that condenses long news articles into short, useful summaries.
• Integrate your custom component into a pipeline, mixing it with existing Haystack modules.

You’ll learn how to go beyond out-of-the-box features and craft components for specialized use cases.

---

4️⃣ Fallback with Branching Pipeline

Sometimes AI systems fail or produce empty results. In this lesson, you’ll learn to handle that gracefully:

• Use branching pipelines to add decision points.
• Implement fallback logic (e.g., if a retriever fails, fall back to a different one).
• Improve system reliability by ensuring robust outputs even in edge cases.

This prepares you for real-world scenarios where robustness is essential.

---

5️⃣ Self-Reflecting Agents with Loops

Agents can think about their own outputs and refine them. In this lesson, you’ll build self-reflecting agents:

• Implement feedback loops where the agent checks its own answers.
• Learn how looping helps reduce hallucinations and improve accuracy.
• Explore advanced agent workflows for complex tasks.

This introduces a more autonomous and intelligent way of building AI agents.

---

6️⃣ Chat Agent with Function Calling

Finally, you’ll bring everything together by building an interactive AI agent:

• Create a chat agent that can have back-and-forth conversations.
• Use function calling to let the agent trigger tools (like search, summarization, or APIs).
• Build an AI that doesn’t just answer questions, but can act dynamically.

This is the capstone skill for building real AI assistants with Haystack.

---

🙌 Acknowledgements

This course and its content are developed by DeepLearning.AI in collaboration with Haystack (deepset). All credits for teaching and materials go to them - this repo simply organizes the notebooks for easy access and learning.

---

✨ With these lessons, you’ll gain the skills to build robust, customized, and intelligent AI applications using Haystack.