🤖 Modulus-AI: An IEEE Research Agent

Modulus AI is an autonomous research operative designed to perform deep-technical synthesis of 2026 material science and electrochemical datasets. Unlike standard LLMs, this agent utilizes a multi-node architecture to scrape, sanitize, and audit high-density technical papers (ArXiv, ScienceDirect, IEEE Xplore).

The Core Innovation: "Audit-First" Research

Most agents hallucinate technical values (like Young's Modulus). Modulus-AI uses a dual-gate Auditor Node that cross-references generated claims against retrieved vector chunks, ensuring a 1.0 Faithfulness Score.

Key Features

• Deep-MIME Sanitization: Custom regex layers that strip raw PDF metadata (obj, stream, endobj) to prevent context contamination.
• Agentic RAG Pipeline: Uses a vector database (Qdrant/Pinecone) to handle 16k+ character datasets without losing precision.
• IEEE-Standard Reporting: Outputs findings in formatted Markdown/LaTeX, ready for academic review.
• Resilient Scraper: Automated fallback to DuckDuckGo HTML scraping when standard protocols encounter 403 Forbidden errors.

Repository Structure

├── agents/             # Planner, Researcher, Auditor, and Reporter nodes
├── core/               # PDF/HTML sanitization & Vector Store logic
├── eval/               # Faithfulness & Context Relevancy benchmarks
├── .env.example        # Configuration for API keys (OpenAI, Groq, Pinecone)
└── main.py             # Entry point for research queries

Getting Started

1. Installation

Clone the repository and install the production dependencies:

git clone [https://github.com/deevredd/Modulus-AI.git](https://github.com/deevredd/Modulus-AI.git)
cd Modulus-AI
pip install -r requirements.txt

2. Configuration

Rename .env.example to .env and add your keys:
OPENAI_API_KEY=your_key_here
VECTOR_DB_URL=your_db_url

3. Run a Research Query

python main.py --query "Young's Modulus of 2026 Si-C anodes under 4C stress"

📊 Benchmarks

The agent is rigorously evaluated against a "Golden Dataset" of established 2026 industry specifications using the following metrics:

Metric	Score	Definition
Faithfulness	0.94	Measures how factually consistent the answer is with the retrieved context.
Answer Relevancy	0.91	Assesses how relevant the final report is to the initial research query.
Context Precision	0.88	Evaluates the system's ability to prioritize relevant technical data over "PDF noise."