An advanced Retrieval-Augmented Generation (RAG) solution that utilizes sophisticated deterministic graph algorithms to handle complex question answering tasks. It enables an autonomous agent to answer non-trivial questions from custom datasets while effectively preventing AI hallucinations。
One-Minute Overview#
Controllable-RAG-Agent is an advanced Retrieval-Augmented Generation solution designed to tackle complex questions that simple semantic similarity-based retrieval cannot solve. This project showcases a sophisticated deterministic graph acting as the "brain" of a highly controllable autonomous agent capable of answering non-trivial questions from your own data.
Core Value: Through multi-step reasoning and adaptive planning mechanisms, it ensures answers are strictly based on provided data while avoiding AI hallucinations, even for queries requiring complex reasoning.
Quick Start#
Installation Difficulty: Medium - Requires Python environment and API key configuration, but the project provides detailed installation steps and usage examples.
# Clone the repository
git clone https://github.com/NirDiamant/Controllable-RAG-Agent.git
cd Controllable-RAG-Agent
# Install dependencies
pip install -r requirements.txt
Core Capabilities#
1. Sophisticated Deterministic Graph Reasoning - Understanding Multi-level Problems#
- Employs deterministic graph algorithms as the agent's "brain" to handle problems requiring complex reasoning
- Can decompose complex queries into manageable subtasks, systematically solving problems that require multi-step reasoning
2. Multi-step Reasoning Capability - Solving Association-Based Problems#
- Breaks down complex queries into manageable subtasks, executing them sequentially while accumulating context
- Can answer complex questions requiring understanding relationships between multiple concepts
3. Adaptive Planning - Dynamically Adjusting Response Strategies#
- Continuously updates plans based on new information, flexibly adjusting response paths
- Can dynamically adjust strategies based on newly acquired information during the response process, improving answer accuracy
4. Hallucination Prevention - Ensuring Answers Based on Facts#
- Uses content verification and context distillation to ensure answers strictly follow provided documents
- Eliminates AI hallucinations, providing reliable, document-based factual answers
5. Performance Evaluation - Comprehensive Quality Assessment#
- Uses Ragas metrics for comprehensive quality assessment including answer correctness, faithfulness, relevance, etc.
- Quantifies system performance through multi-dimensional metrics, continuously optimizing answer quality