Adaptive Agentic RAG Pipeline

✨ Adaptive AI assistant with RAG + web search fallback + agentic routing 🌐📚

---

🚀 Overview

RAGent is an advanced Adaptive RAG (Retrieval-Augmented Generation) pipeline that intelligently decides how to answer user queries using:

• Internal document retrieval 📚
• Agentic LLM reasoning 🤖
• Conditional routing for unknown queries 🔄
• Web search fallback 🌐

It is designed to mimic human decision-making in multi-step reasoning pipelines and integrates seamlessly with multiple LLM providers via OpenRouter.

---

🧩 Features

• Adaptive RAG: Dynamically retrieves and ranks relevant documents.
• Agentic Control Flow: Conditional routing to select between internal knowledge or web search.
• Web Search Fallback: Automatically queries the web if internal knowledge is insufficient.
• Multi-LLM Integration: Supports OpenRouter, GPT models, and other LLMs.
• Prompt Engineering: Modular and reusable prompt templates for different contexts.
• Message Conversion: Converts prompts into LLM-compatible messages.
• Monitoring Ready: Easily extendable to track latency, costs, and evaluation metrics.

---

🏗️ Architecture

flowchart LR
    User[User Query] --> Retriever[BM25 Retriever]
    Retriever --> PromptBuilder[Prompt Builder]
    PromptBuilder --> MessageConverter[ChatMessageConverter]
    MessageConverter --> LLM[OpenRouter LLM Generator]
    LLM --> Router[Conditional Router]
    Router -->|answer| Output[Return Answer]
    Router -->|go_to_websearch| WebSearch[SerperDev Web Search]
    WebSearch --> WebPrompt[Web Prompt Builder]
    WebPrompt --> WebMessageConverter[ChatMessageConverter]
    WebMessageConverter --> WebLLM[OpenRouter Web Generator]
    WebLLM --> Output

Loading