Advent of Haystack

A collection of Haystack-powered projects based on Advent of Haystack. View my original solutions here.

What is Haystack?

Haystack is an open-source framework by deepset for building:

• Production-ready LLM applications
• Retrieval-augmented generative pipelines
• State-of-the-art search systems for large document collections

Overview

This repository contains 10 projects demonstrating pipeline building with Haystack and integrations with:

• Weaviate - Vector Database
• AssemblyAI - Speech AI
• NVIDIA - GPU Computing
• Arize - ML Observability
• MongoDB - Document Database

Folder Structure

• /: Jupyter notebooks containing experiments and implementations

Projects

1. Web Knowledge Retrieval System

• Fetches content from URLs
• Ranks documents by relevance
• Identifies top 10 relevant documents
• Enables seamless querying and answering

2. RAG Pipeline with Weaviate

• Implements Retrieval-Augmented Generation with vector database
• Combines vector search with structured filtering
• Stores both objects and vectors efficiently

3. Multi-Query Retrieval Pipeline

• Implements custom Haystack component
• Generates multiple queries from single input
• Handles parallel query processing

4. Audio Transcription Pipeline

• Integrates AssemblyAI for speech processing
• Performs speech-to-text conversion
• Enables speaker diarization
• Generates responses based on transcriptions

5. RAG Development with deepset Studio

• Utilizes UI-based interface of deepset Studio for pipeline creation
• Leverages deployment tools and templates

6. NVIDIA Inference Microservice Architecture

• Integrates with NVIDIA inference microservices for ranking and embeddings
• Implements task delegation optimization
• Enables multilingual embedding document grouping

7. Intelligent Toy Matching System & LLM Monitoring with Arize Phoenix

• Creates personalized toy recommendations
• Integrates Arize Phoenix for:
  • Real-time performance tracking
  • LLM response evaluation
  • Trace data visualization
  • Quality monitoring dashboards
• Implements automated evaluation pipeline

8. Inventory Management Agent with Custom Tools

• Manages inventory tracking system
• Implements RAG-based web search
• Provides price checking functionality
• Enables CRUD operations on inventory
• Uses DuckDuckGo integration for web search

9. Self-Reflecting Gift Recommendation System

• Leverages MongoDB Atlas vector search capabilities
• Enhances semantic search using OpenAI embeddings
• Generates personalized gift recommendations via GPT models
• Implements advanced reasoning with self-reflection
• Streamlines gift metadata management and organization

10. Evaluating a RAG Pipeline with EvaluationHarness

• Uses HotpotQA dataset for testing
• Implements multiple evaluation metrics
• Integrates with Hugging Face and OpenAI
• Provides comprehensive pipeline assessment
• Enables model comparison and parameter tuning