Orchestrix - Multi-Agent Research Intelligence Platform

Orchestrix is a full-stack web application that eliminates fragmentation in the academic research workflow through multi-agent AI orchestration. It queries multiple academic databases, performs analysis, generates citations, provides synthesis capabilities, detects conflicts between agent outputs, and supports scheduled research digests.

Prerequisites

• Python 3.11+
• Node.js 18+
• npm or yarn

Setup

1. Clone the Repository

git clone <repository-url>
cd Evolve-Hackathon

2. Backend Setup

cd backend

# Create virtual environment (recommended)
python3.12 -m venv venv
source venv/bin/activate  # On Windows: venv\Scripts\activate

# Install dependencies
pip install -r requirements.txt

# Copy environment file and add your API key
cp .env.example .env
# Edit .env and add your GEMINI_API_KEY

# The application will automatically create the SQLite database on first run

3. Frontend Setup

cd ../frontend

# Install dependencies
npm install

# Create environment file
echo "VITE_API_BASE_URL=http://localhost:8000" > .env

4. Running the Application

Terminal 1 - Backend:

cd backend
source venv/bin/activate
uvicorn main:app --reload --host 0.0.0.0 --port 8000

Terminal 2 - Frontend:

cd frontend
npm run dev

The application will be available at:

• Frontend: http://localhost:5173
• Backend API: http://localhost:8000
• API Documentation: http://localhost:8000/docs

Architecture

Backend (Python FastAPI)

backend/
├── main.py              # FastAPI entrypoint and REST endpoints
├── orchestrator.py      # Central coordination layer
├── scheduler.py         # Background task scheduler for digests
├── models.py            # SQLAlchemy database models
├── database.py          # Database session management
├── schemas.py           # Pydantic request/response models
├── requirements.txt     # Python dependencies
├── .env.example         # Environment variables template
└── agents/
    ├── discovery.py      # Research Discovery Agent
    ├── analysis.py       # Analysis Agent
    ├── citation.py       # Citation Generator Agent
    ├── summarizer.py     # Summarization Agent
    ├── conflict_detector.py  # Conflict Detection Agent
    └── digest_scheduler.py   # Scheduled Digest Agent

Frontend (React + Vite)

frontend/
├── package.json
├── vite.config.js
├── index.html
└── src/
    ├── main.jsx
    ├── App.jsx
    ├── api.js
    ├── pages/
    │   ├── Search.jsx
    │   ├── Dashboard.jsx
    │   └── SessionCompare.jsx
    └── components/
        ├── AgentTraceLog.jsx
        ├── PaperCard.jsx
        ├── AnalysisCharts.jsx
        ├── CitationPanel.jsx
        ├── SummaryPanel.jsx
        └── SessionSidebar.jsx

Database Schema

Sessions Table

Column	Type	Description
id	UUID	Primary key
name	String	User-defined session name
query	String	Original search query
created_at	DateTime	UTC timestamp
updated_at	DateTime	Last update timestamp

Papers Table

Column	Type	Description
id	UUID	Primary key
session_id	UUID	Foreign key to sessions
title	String	Paper title
authors	JSON	Array of author names
year	Integer	Publication year
abstract	Text	Paper abstract
source_url	String	URL to the paper
citation_count	Integer	Number of citations
relevance_score	Float	Computed relevance score
external_id	String	ID from source API
source	String	"semantic_scholar" or "arxiv"

Analyses Table

Column	Type	Description
id	UUID	Primary key
session_id	UUID	Foreign key to sessions
analysis_type	String	One of: publication_trend, top_authors, keyword_frequency, citation_distribution, emerging_topics
data_json	JSON	Computed analysis results

Summaries Table

Column	Type	Description
id	UUID	Primary key
paper_id	UUID	Foreign key to papers
abstract_compression	Text	Plain-language summary
key_contributions	Text	Main novel contributions
methodology	Text	Methods/approach description
limitations	Text	Known/inferred limitations

Citations Table

Column	Type	Description
id	UUID	Primary key
paper_id	UUID	Foreign key to papers
apa	Text	APA format citation
mla	Text	MLA format citation
ieee	Text	IEEE format citation
chicago	Text	Chicago format citation

Notes Table

Column	Type	Description
id	UUID	Primary key
paper_id	UUID	Foreign key to papers
content	Text	Free-text notes

Conflicts Table

Column	Type	Description
id	UUID	Primary key
session_id	UUID	Foreign key to sessions
conflict_type	String	Type: semantic_contradiction, methodology_mismatch, temporal_contradiction, etc.
severity	String	high, medium, or low
title	String	Brief conflict title
description	Text	Detailed description
analysis_insight	Text	What the Analysis Agent found
summarization_insight	Text	What the Summarization Agent found
resolved	Boolean	Whether conflict is resolved
resolution_notes	Text	Notes on resolution

Scheduled Digests Table

Column	Type	Description
id	UUID	Primary key
name	String	Digest name
query	String	Search query to run
frequency	String	daily, weekly, biweekly, or monthly
last_run_at	DateTime	When digest last ran
next_run_at	DateTime	When digest will next run
is_active	Boolean	Whether digest is active
notify_email	String	Email for notifications

Digest Runs Table

Column	Type	Description
id	UUID	Primary key
scheduled_digest_id	UUID	Foreign key to scheduled_digests
session_id	UUID	Foreign key to sessions (created session)
query	String	Query that was run
new_papers_count	Integer	Number of new papers found
new_paper_ids	JSON	Array of new paper external IDs
status	String	pending, running, completed, or failed
error_message	Text	Error details if failed

Relevance Scoring Formula

Papers are ranked using a weighted combination of three factors:

score = 0.5 × normalize(citation_count) + 0.3 × normalize(year) + 0.2 × keyword_match_ratio

Where:

• normalize(citation_count): Scales citations between 0-1 relative to the min/max in the result set
• normalize(year): Scales year between 0-1 relative to min/max (with min=1990)
• keyword_match_ratio: Counts query words appearing in title+abstract, divided by total query words (stopwords removed)

Safety Score Algorithm

The Safety Score feature is not applicable to this application as it focuses on academic research discovery rather than content safety assessment.

Multi-Agent Pipeline

Agent 1: Discovery Agent

Queries Semantic Scholar and arXiv APIs in parallel, normalizes results, deduplicates by title similarity, and computes relevance scores.

Agent 2: Analysis Agent

Generates five types of analysis:

• Publication Trend: Papers per year
• Top Authors: Top 15 authors by frequency
• Keyword Frequency: Top 40 keywords from abstracts
• Citation Distribution: Histogram buckets (0, 1-10, 11-50, 51-200, 201-1000, 1000+)
• Emerging Topics: Words with highest delta between recent (2020+) and historical frequency

Agent 3: Citation Generator Agent

Uses Gemini API to generate properly formatted citations in APA, MLA, IEEE, and Chicago styles.

Agent 4: Summarization Agent

Two functions:

• summarize_paper: Generates structured summary with abstract compression, key contributions, methodology, and limitations
• synthesize_papers: Combines multiple papers into a cohesive paragraph identifying common themes, contradictions, and research gaps

Agent 5: Conflict Detection Agent

Detects contradictions between Analysis and Summarization agents:

• Semantic Contradiction: Analysis keywords vs summary content
• Methodology Mismatch: Analysis focus vs paper methodology
• Temporal Contradiction: Emerging topics vs historical references
• Citation Impact Contradiction: High citations vs perceived limitations
• Author Dominance Contradiction: Few authors dominate vs underemphasized contributions
• Keyword Mismatch: Analysis keywords vs summary emphasis

Agent 6: Digest Scheduler Agent

Manages scheduled research digests:

• Supports daily, weekly, biweekly, and monthly frequencies
• Tracks papers seen to identify truly new papers
• Generates email notifications for new findings
• Creates sessions for digest runs

API Endpoints

Core Endpoints

Method	Endpoint	Description
GET	/health	Health check
POST	/sessions	Create new session
GET	/sessions	List all sessions
GET	/sessions/{id}	Get session with all data
POST	/sessions/{id}/orchestrate	Run orchestration pipeline
PATCH	/papers/{id}/note	Update paper note
POST	/sessions/{id}/synthesize	Synthesize selected papers
GET	/sessions/{id}/export/bib	Export as BibTeX
GET	/sessions/{id}/export/txt	Export citations as text

Conflict Resolution Endpoints

Method	Endpoint	Description
GET	/sessions/{id}/conflicts	Get all conflicts for a session
POST	/sessions/{id}/conflicts/{conflict_id}/resolve	Mark conflict as resolved
POST	/sessions/{id}/detect-conflicts	Run conflict detection on session

Scheduled Digest Endpoints

Method	Endpoint	Description
POST	/digests	Create scheduled digest
GET	/digests	List all scheduled digests
GET	/digests/{id}	Get digest with run history
DELETE	/digests/{id}	Delete scheduled digest
PATCH	/digests/{id}/toggle	Enable/disable digest
POST	/digests/{id}/run	Manually trigger digest run
GET	/digests/{id}/preview	Preview new papers for digest

External APIs Used

Semantic Scholar API

• Purpose: Academic paper search and metadata
• Documentation: https://api.semanticscholar.org/
• Rate Limits: Varies by tier

arXiv API

• Purpose: Open-access preprint repository
• Documentation: https://arxiv.org/help/api
• Rate Limits: 1 request per 3 seconds

Google Gemini API

• Purpose: AI-powered citation formatting and summarization
• Documentation: https://ai.google.dev/docs
• Requirements: API key in .env file

Environment Variables

Backend (.env)

GEMINI_API_KEY=your_gemini_api_key_here
DATABASE_URL=sqlite:///./orchestrix.db
FRONTEND_URL=http://localhost:5173

Frontend (.env)

VITE_API_BASE_URL=http://localhost:8000

Features

Search Page

• Prominent search bar for research queries
• Live agent activity trace with status indicators
• Four tabs: Papers, Analysis, Citations, Summary
• Paper cards with relevance score badges
• Interactive Recharts visualizations

Dashboard Page

• Left sidebar with past sessions
• Full session details with all data
• Auto-saving notes with debounced updates
• Tab-based navigation

Compare Page

• Side-by-side session selection
• Merged publication trend charts
• Unique keyword highlighting
• Paper overlap detection

Conflict Resolution

• Automatic conflict detection during orchestration
• Manual conflict detection for existing sessions
• Conflict categorization by type and severity
• Resolution tracking with notes

Scheduled Research Digests

• Create recurring queries with customizable frequency
• Automatic detection of new papers
• Email notifications for new findings
• Historical run tracking and statistics

License

MIT License