Research Assistant System (RAS)

A multi-agent AI system that helps researchers write better academic papers by providing real-time feedback, literature discovery, and intelligent Q&A.

Features

• Writing Corrections - Grammar, style, and logical consistency checks
• Literature Support - Finds supporting evidence and opposing viewpoints from indexed papers
• Research Chatbot - Answers questions with citations from the paper corpus

Architecture

User → Interface Agent → Router Agent → Scout Agent → Expert Agents
                                              ↓
                                     Pinecone (Hybrid Search)

Agent	Role
Interface	Analyzes paragraphs for grammar, logic, and facts needing verification
Router	Orchestrates queries between agents, manages expert registry
Scout	Retrieves chunks via hybrid search, spawns experts when needed
Expert	Deep Q&A on a specific paper (one expert per paper)

Prerequisites

• Python 3.12+
• MongoDB (for agent checkpointing)
• Terraform 1.5+ (for infrastructure)
• AWS CLI configured

Quick Start

1. Install Dependencies

uv sync

2. Configure Environment

Create a .env file:

OPENAI_API_KEY=sk-xxx
PINECONE_API_KEY=xxx
MONGO_URI=mongodb://localhost:27017
BOOTSTRAP_BUCKET_NAME=your-bucket
PAPERS_S3_PREFIX=development/your-username/papers

3. Run the Application

uvicorn app.main:app --reload

API docs at http://localhost:8000/docs

API Endpoints

Method	Endpoint	Description
POST	/api/v1/conversations	Create new conversation
GET	/api/v1/conversations/{id}	Get conversation
DELETE	/api/v1/conversations/{id}	Delete conversation
POST	/api/v1/analyze	Analyze a paragraph
GET	/api/v1/health	Health check

Example Usage

# Create conversation
curl -X POST http://localhost:8000/api/v1/conversations

# Analyze paragraph
curl -X POST http://localhost:8000/api/v1/analyze \
  -H "Content-Type: application/json" \
  -d '{
    "paragraph": "The attention mechanism allows models to focus on relevant parts of the input.",
    "paragraph_index": 0,
    "conversation_id": "<id>"
  }'

Infrastructure Setup

Bootstrap (Admin)

python setup.py bootstrap init
python setup.py bootstrap apply
aws iam add-user-to-group --group-name ras-developers --user-name {username}

Dev Environment

$env:UPSTASH_EMAIL = "your-email"
$env:UPSTASH_API_KEY = "your-api-key"

python assume_role.py | Invoke-Expression
python setup.py dev init
python setup.py dev apply

Ingestion Pipeline

The pipeline indexes papers into Pinecone for retrieval:

# Producer: Fetch papers from ArXiv → S3 → SQS
python run.py producer up

# Consumer: Parse PDFs → chunk → embed → Pinecone
python run.py consumer up

Mode	Description
parse	Parse PDFs, save raw elements to S3
process	Chunk, embed, store to Pinecone
full	Both (default)

CI/CD

Branch	Environment
main	Staging
prod	Production

GitHub Secrets required:

• OPENAI_API_KEY
• PINECONE_API_KEY
• UNSTRUCTURED_API_KEY
• UPSTASH_EMAIL / UPSTASH_API_KEY

Documentation

See dev_docs.md for detailed documentation:

• Multi-agent implementation details
• Terraform infrastructure
• Retrieval service configuration
• Evaluation results