AI Research Assistant

A LangGraph-powered AI Research Assistant designed to automate the process of academic research. It creates plans, queries academic databases, downloads/ingests PDFs, performs RAG (Retrieval-Augmented Generation) using a hybrid vector store, and generates comprehensive comparative reports.

Deploy

---

Features

• Agentic LangGraph Workflow: Autonomous orchestration of research plans, web searches, paper downloads, ingestion, and multi-step report generation.
• Hybrid Retrieval (RAG): Combines OpenAI dense embeddings (text-embedding-3-small) with BM25 sparse embeddings (FastEmbedSparse) in a Qdrant vector database.
• Academic Search & Ingestion: Searches arXiv and general web databases, downloads papers as PDFs, and chunks/stores them automatically.
• Streaming Web Interface: Simple Flask frontend that displays live updates and logs from the LangGraph execution using Server-Sent Events (SSE).
• Structured Reports: Produces structured Markdown reports covering research findings, paper comparisons, and trend analysis.

---

Prerequisites

• Python 3.11+
• Qdrant Vector Database (or Docker to run it)
• API Keys: OpenAI API key (for embeddings) and DeepSeek API key (for the LLM).

---

Installation & Setup

1. Clone the Repository

git clone https://github.com/trungbb7/Research-Assistant.git
cd ai-agent

2. Set Up Virtual Environment & Dependencies

python -m venv venv
# On Windows
venv\Scripts\activate
# On macOS/Linux
source venv/bin/activate

pip install -r requirements.txt

3. Configure Environment Variables

Create a .env file in the root directory:

DEEPSEEK_API_KEY=your_deepseek_api_key
OPENAI_API_KEY=your_openai_api_key
LLM_PLATFORM=OPENAI
QDRANT_HOST=localhost
QDRANT_PORT=6333

---

Running the Application

Option A: Running Locally

1. Start Qdrant Vector DB (via Docker):

   docker run -d -p 6333:6333 -p 6334:6334 qdrant/qdrant

2. Start the Flask Application:

   python app.py

   Open your browser and navigate to http://localhost:5000.

---

Option B: Running with Docker Compose

To spin up both the Flask application and the Qdrant instance together:

docker-compose up --build

Access the application at http://localhost:5000.