Leveraging LangChain and Hugging Face Transformers, this project implements a Retrieval-Augmented-Generation (RAG) system designed to optimize LLM-generated responses. The system aims to create an intelligent agent capable of multi-turn conversations with integrated memory for enhanced context retention.
- Results for various stages of running can be found
reports/outputs/
- Supplement learning through project-based experience while completing NVIDIA's
Building RAG Agents with LLMscourse. - Improve proficiency with LangChain and modern methods of document retrieval, embedding, and augmentation.
- Utilise a Retrieval-Augmented-Generation (RAG) pipeline to enhance knowledge of generative AI and its practical applications.
- Build a modular architecture that allows easy integration and replacement of LLMs, including more advanced models like OpenAI’s ChatGPT or other future offerings.
- Python 3.12.8
- LangChian
- HuggingFace transformers
- FAISS Vectorstore
- Gradio
- Research papers, both local .pdf files and requested from Arxiv via API
- https://arxiv.org/abs/2501.08476 (citation - arXiv:2501.08476 [quant-ph])
- https://arxiv.org/abs/2501.09079 (citation - arXiv:2501.09079 [quant-ph])
- Document Retrieval and Augmentation: Built retrieval systems using FAISS to identify and retrieve relevant document chunks, leveraging their context for response generation. Augmented the retrieved data with additional metadata for improved query results.
- Conversational Memory: Integrated memory-based retrieval pipelines using LangChain's ConversationalRetrievalChain and ConversationBufferMemory to maintain multi-turn conversational context.
- LLM Integration: Utilised Hugging Face Transformers for text generation, deploying a local model pipeline to ensure efficient and contextually accurate responses.
- Pipeline Modularity: Designed a modular RAG pipeline architecture with separate components for retrieval, response generation, and user interaction, ensuring scalability and maintainability.
- User Interaction: Developed a Gradio-based conversational dashboard for real-time interactions with the RAG pipeline, enabling seamless multi-turn conversations with source document attribution.
- Data Handling: Implemented a robust data handling system using LangChain's utilities and custom DataModule classes for managing vector stores, retrieval pipelines, and conversational chains.
git clonethe repo andcdinto the directory- Create virtual environment and install dependencies
pip install -r requirements.txt - Run the pipeline
python main.py - Launch localgradio interface to view conversational dashboard
Utilised soley as a source of raw data, for testing/demonstrative purposes.
- arXiv:2501.08476 [quant-ph]
- arXiv:2501.09079 [quant-ph]
For more information, or to discuss this project further, please reach out.
Email: delstonds@outlook.com