🤖 AgenticRAG: Adaptive & Self-Correcting RAG System

AgenticRAG is a sophisticated Retrieval-Augmented Generation (RAG) system built on LangGraph. It implements the latest "Agentic RAG" patterns, including Adaptive Routing and Corrective Self-Healing, ensuring high-accuracy answers by grading its own work.

---

🚀 Key Advanced Features

🔄 Adaptive Routing

The system doesn't assume all answers are in the vector store. It uses an LLM-powered router to decide whether a query should be served by the Local Vector Store (domain-specific knowledge) or Web Search (real-time/general info via Tavily).

🔎 Corrective Retrieval (Self-Grading)

Retrieved documents are automatically graded for relevance. If the system finds the retrieved data is insufficient or irrelevant:

1. It transforms the query to better capture intent.
2. It re-tries the retrieval or falls back to web search.

🛡️ Hallucination & Answer Grading

Before a response is shown to the user, CodeWeaver performs two critical checks:

1. Hallucination Check: Is the answer supported only by the retrieved facts?
2. Utility Check: Does the answer actually resolve the user's original question?

---

🛠️ Tech Stack

• Orchestration: LangGraph
• Vector Store: FAISS
• Embeddings: HuggingFace (all-MiniLM-L6-v2)
• LLM: Groq (Llama 3.3 70B)
• Tools: Tavily Search API

---

📈 The Agentic Workflow

graph TD
    START --> Router[Router Node]
    Router -->|Local| Retrieve[Retrieve from VectorStore]
    Router -->|General| WebSearch[Web Search]
    Retrieve --> Grade[Grade Documents]
    Grade -->|Relevant| Generate[Generate Answer]
    Grade -->|Irrelevant| Transform[Transform Query]
    Transform --> Retrieve
    WebSearch --> Generate
    Generate --> Hallucination{Hallucination Check}
    Hallucination -->|Failed| Generate
    Hallucination -->|Passed| AnswerCheck{Answer Utility Check}
    AnswerCheck -->|Not Useful| Transform
    AnswerCheck -->|Useful| END

Loading

---

🚦 Getting Started

1. Requirements

• Python 3.9+
• Groq API Key
• Tavily API Key

2. Installation

pip install -r requirements.txt

3. Ingestion

Create your vector index first:

python ingest.py

4. Usage

python AdaptiveCorrectiveAgenticRag.py

---

📂 Project Structure

• AdaptiveCorrectiveAgenticRag.py: The main graph logic.
• ingest.py: Data ingestion and vector store creation.
• faiss_index/: Local vector database.
• requirements.txt: Project dependencies.

---

Created as part of the LangGraph Learning Series - Tier 1 Showcase.