Research AI Assistant Application - RAG

This is a submission for the Open Source AI Challenge with pgai and Ollama

What I Built

I created an AI-powered research assistant application designed to make finding and understanding relevant research papers much easier. When you’re writing a research paper, it’s often a struggle to locate high-quality papers that match your topic closely. This app streamlines that process. It first searches and pulls up research papers closely aligned with your query, and then leverages LLM (Large Language Models) and RAG (Retrieval-Augmented Generation) to generate insightful summaries, suggested responses, and information based on your specific prompt. Not only does it provide summaries, but it also includes links, titles, and other essential details for each recommended paper, so you can get a comprehensive view of what's available. This tool essentially simplifies the process of gathering and interpreting research, helping you focus more on writing and analysis than on the hunt for resources.

Key Features : -

1. Relevant Paper Discovery: Quickly finds research papers closely related to your topic or query.
2. Intelligent Summarization: Uses Large Language Models (LLMs) and Retrieval-Augmented Generation (RAG) to generate clear, concise summaries for each paper.
3. Prompt-Based Response Generation: Provides insightful responses and context-specific information based on your prompt, helping you understand complex topics.
4. Link and Title Integration: Includes direct links, titles, and summaries of recommended papers, saving you time and ensuring easy access to source material.
5. User-Friendly Interface: Simplifies the research process, allowing you to focus on writing and analysis rather than sourcing information.

Demo

Here is my Github Link : - https://github.com/Ayush-developer/Reasearch_Assistant_RAG_Application

Tools Used

• Timescale Cloud: Database hosting
• pgai through TimescaleDB for AI
• pgVector: For efficient data storage and retrieval of embeddings
• Ollama: For generating embeddings and handling response generation using Mistral and LLama
• React.js: Used for building the user interface
• Flask API ; - For backend API Handling
• Postman : - Testing Queries

Technologies Used : -

1. Ollama with Nomic Embeddings: Utilizes Nomic embeddings for accurate text representation, supporting semantic search and relevance scoring for research queries.

2. PostgreSQL with PG Vector and Psycopg2: Stores embeddings in PostgreSQL using pgvector, enabling efficient similarity searches, with psycopg2 managing database connections.

3. Llama Model via Ollama: Leverages the Llama model for natural language processing, generating precise responses and summaries based on user queries.

4. Flask API for Backend: Manages backend processing and data flow between the frontend and AI models, ensuring smooth interaction.

5. React for Frontend: Provides a responsive, intuitive interface for query input, viewing results, and accessing summaries.

Final Thoughts

This is my first time participating in a Dev.to Hackathon, and it’s been an incredible learning experience! Building this AI application has opened my eyes to so many new technologies, from working with Large Language Models and embeddings to managing backend integrations. I’m still figuring out deployment and improving the application’s functionality, so I’d love to connect with anyone willing to help or share advice. As a 4th-year student currently interning with React, this project marked my first foray into advanced AI and NLP, and I’m grateful for the opportunity to showcase my work and grow through this experience. Thank you for creating such an encouraging platform!

Prize Categories:
Open-source Models from Ollama: AI Bot leverages Ollama’s open-source models for embedding and understanding text-based queries.