CYC2002tommy/Deep-Research-Agent

What it solves

Deep Research Agent (Deep Science Writer) automates the tedious and error-prone process of academic literature reviews. It eliminates common AI issues like hallucinated citations, "AI fluff" in academic writing, and the struggle of gathering full-text papers from restricted databases.

How it works

The project implements a 7-phase autonomous pipeline designed for the Hermes/ECC framework. It uses a multi-agent system to query databases like Scopus, OpenAlex, and Semantic Scholar. To prevent hallucinations, it follows a strict sequence: it screens abstracts first, then downloads and reads the full text of highly relevant papers to verify claims. It includes a dedicated "anti-hallucination" phase where it pings every DOI to ensure the link is valid. Finally, it uses Python scripts to generate a formatted APA 7th .docx file and integrates with Obsidian and Google NotebookLM for knowledge management.

Who it’s for

Academic researchers and scientists who need a rigorous, evidence-backed literature review process that adheres to strict journal quality standards (Q1-Q2 journals) and avoids the same pitfalls as generic LLMs.

Highlights

• Full-Text Verification: Mandatory reading of full-text PDFs to prevent abstract-induced hallucinations.
• Strict Quality Control: Hard-coded to prioritize Q1-Q2 journals and ban specific publishers like MDPI.
• DOI Validation: Automatically tests every generated DOI via HTTP requests to ensure 100% validity.
• Academic Tone Enforcement: Uses a specialized peer-review loop to strip generic AI vocabulary and enforce a professional tone.
• Hands-Free Output: Programmatically generates Microsoft Word documents with APA 7th formatting and data visualizations.
• Knowledge Loop: Automatically syncs research summaries to Obsidian and uploads sources to Google NotebookLM.