🔬 ScholarAgent — Autonomous AI Scientific Research Agent

Try the Live Interactive Demo: https://kushagracancode.github.io/autonomous-research-agent/

---

🎯 System Overview

ScholarAgent is a production-grade autonomous multi-agent scientific literature research system. Given a natural language research inquiry:

$$\text{"What are the latest methods for detecting AI-generated deepfakes?"}$$

ScholarAgent autonomously plans, searches, reads, cross-examines, and synthesizes publication-grade literature reports with zero human intervention:

$$\text{User Inquiry} \longrightarrow \text{Planning} \longrightarrow \text{arXiv Search} \longrightarrow \text{Evidence Extraction} \longrightarrow \text{Hallucination Verification} \longrightarrow \text{Gap Synthesis} \longrightarrow \text{Publication Report}$$

---

🏛️ Multi-Agent Architecture

                       ┌───────────────────────┐
                       │ User Research Inquiry │
                       └───────────┬───────────┘
                                   │
                                   ▼
                       ┌───────────────────────┐
                       │   1. Planner Agent    │
                       │ (Sub-queries & Scope) │
                       └───────────┬───────────┘
                                   │
                    ┌──────────────┴──────────────┐
                    ▼                             ▼
         ┌─────────────────────┐       ┌─────────────────────┐
         │ 2. arXiv Retriever  │       │ Crossref / Semantic │
         │ (Live Academic API) │       │   Scholar Search    │
         └──────────┬──────────┘       └──────────┬──────────┘
                    └──────────────┬──────────────┘
                                   │
                                   ▼
                       ┌───────────────────────┐
                       │ 3. Evidence Extractor │
                       │ (Methods, Datasets)   │
                       └───────────┬───────────┘
                                   │
                                   ▼
                       ┌───────────────────────┐
                       │ 4. Verification Guard │
                       │ (Hallucination Audit) │
                       └───────────┬───────────┘
                                   │
                                   ▼
                       ┌───────────────────────┐
                       │ 5. Gap Synthesizer    │
                       │ (Systemic Bottlenecks)│
                       └───────────┬───────────┘
                                   │
                                   ▼
                       ┌───────────────────────┐
                       │ 6. Publication Writer │
                       │ (Markdown / PDF / Bib)│
                       └───────────────────────┘

---

🔬 Core Capabilities

1. 🧠 Dynamic Research Planning (engine/planner.py)

• Analyzes inquiry domain taxonomy.
• Formulates 5 structured scientific sub-questions.
• Generates 4-6 targeted, high-precision Boolean search queries optimized for the arXiv search grammar.

2. 📄 Live arXiv Academic Retrieval (engine/retriever.py)

• Direct live interface with the arXiv API (Atom/XML feed parsing).
• Deduplicates papers, extracts author lists, publication dates, and abstracts.
• Computes multi-term semantic relevance ranking.

3. 🔬 Evidence & Metric Extraction (engine/extractor.py)

• Classifies research methodologies (Frequency-Domain Spectral Fingerprinting, Biological & rPPG Coherence, Spatiotemporal Vision Transformers, Diffusion Artifact Analysis).
• Extracts benchmark datasets (FaceForensics++, Celeb-DF, WildDeepfake, DFDC).
• Isolates quantitative metrics ($97.8%$ AUC, $99.1%$ AP, $3.8%$ EER).

4. 🛡️ Hallucination Detection & Factuality Scoring (engine/verifier.py)

• Every generated claim is cross-examined against primary source passages.
• Assigns confidence tiers:
  • 🟢 Supported ($\ge 85%$ vocabulary and numerical alignment).
  • 🟡 Partially Supported ($50% - 84%$ alignment).
  • 🔴 Unsupported ($< 50%$ alignment; flagged as generative drift).
• Produces a real-time Factuality Index and Hallucination Rate.

5. 💡 Research Gap & Hypothesis Detection (engine/gap_analyzer.py)

• Compares cross-architecture tradeoffs.
• Isolates common failure modes (e.g. cross-dataset generalization drops, lossy social media compression erasure, quadratic temporal latency).
• Formulates novel, actionable scientific research opportunities.

6. ✍️ Publication-Grade Synthesis (engine/writer.py)

• Complete executive abstract.
• Comparative benchmark matrix table.
• Thematic literature review.
• Evidentiary grounding audit.
• Hyperlinked IEEE / APA academic bibliography.

---

💻 Quickstart & Local Setup

Prerequisites

• Python 3.10+
• Internet connection (for live arXiv API queries)

Installation

# 1. Clone repository
git clone https://github.com/KushagraCanCode/autonomous-research-agent.git
cd autonomous-research-agent

# 2. Install dependencies
pip install fastapi uvicorn requests pydantic jinja2

# 3. Launch the FastAPI server
python -m uvicorn backend.main:app --port 8100 --reload

# 4. Open in browser:
# http://localhost:8100

---

🌐 Public Web Access

Anyone can explore and test ScholarAgent without installing any dependencies:

👉 https://kushagracancode.github.io/autonomous-research-agent/