ResearchFlow

A production-grade multi-agent research system powered by LangGraph. Searches the web, reads documents, synthesizes findings, and iteratively refines results—all autonomous, all automated.

Features

• Multi-Agent Orchestration — Web search, document parsing, summarization, and critique agents work together
• Reflection Loop — Automatically refines low-quality summaries (max 3 iterations)
• Citation Tracking — Every claim is traceable back to source URLs
• Fast LLM — Powered by Groq's 70B model (inference in <2s)
• Interactive CLI — Persistent chat for multi-turn research
• Quality Evaluation — Built-in golden set testing framework

Architecture

graph TD
    A["Research Question"] --> B["Web Search Agent\n(Perplexity API)"]
    B --> C["Document Reader Agent\n(BeautifulSoup)"]
    C --> D["Summarizer Agent\n(Groq LLM)"]
    D --> E["Critic Agent\n(Groq LLM)"]
    E --> F{Score < 7?}
    F -->|YES\nMax 3x| B
    F -->|NO| G["Finalize Report\n(with Citations)"]
    
    style A fill:#3498db,stroke:#2c3e50,color:#fff
    style B fill:#2ecc71,stroke:#2c3e50,color:#fff
    style C fill:#9b59b6,stroke:#2c3e50,color:#fff
    style D fill:#e74c3c,stroke:#2c3e50,color:#fff
    style E fill:#f39c12,stroke:#2c3e50,color:#fff
    style F fill:#34495e,stroke:#2c3e50,color:#fff
    style G fill:#1abc9c,stroke:#2c3e50,color:#fff

Loading

Tech Stack

• LangGraph — Agent orchestration and routing
• Perplexity API — Real-time web search with citations
• Groq API — Fast LLM for synthesis (llama-3.3-70b)
• BeautifulSoup + Requests — Document parsing
• Python 3.11+

Quick Start

1. Clone and Install

git clone <repo>
cd research-assistant
python -m venv venv
source venv/bin/activate
pip install -r requirements.txt

2. Get API Keys

All free/generous tiers:

• Perplexity: https://www.perplexity.ai/settings/api → Copy key to .env as PERPLEXITY_API_KEY
• Groq: https://console.groq.com/keys → Copy key to .env as GROQ_API_KEY
• LangSmith (optional): https://smith.langchain.com → Copy key to .env as LANGSMITH_API_KEY

3. Configure

cp .env.example .env
# Edit .env with your API keys

Usage

python main.py

Real example output from Perplexity + Groq

ResearchFlow
Type your research question and press Enter. Type 'exit' to quit.

> What are the latest breakthroughs in quantum computing?

Researching...

Final Report:

# Research Report

## Question
What are the latest breakthroughs in quantum computing?

## Summary
The landscape of quantum computing has witnessed significant breakthroughs as of 
early 2026, with advancements in areas such as Majorana qubit readout, scalable 
ion-trap systems, error correction in superconducting qubits, and the development 
of molecular qubits for quantum networks [1][2][3][5]. One of the key milestones 
was the experimental readout of a Majorana qubit using quantum capacitance, which 
enables millisecond coherence times and reduces error rates [1]. Furthermore, 
Quantinuum's Helios system, launched in November 2025, is claimed to be the most 
accurate quantum computer yet, capable of performing quantum AI tasks that would 
require immense classical computational power [2]. Google's Quantum Echoes 
algorithm, announced in October 2025, demonstrates verifiable quantum advantage 
on the Willow chip, computing molecular structures significantly faster than 
classical supercomputers [3].

## Evaluation
Score: 8/10

Feedback: The summary provides a comprehensive overview of the latest breakthroughs 
in quantum computing, covering significant advancements in areas such as Majorana 
qubits, scalable ion-trap systems, and molecular qubits. The inclusion of specific 
examples adds depth to the summary.

## Citations

1. Majorana Qubit Readout and Scalable Ion-Trap Systems
   URL: https://www.perplexity.ai
   Snippet: Breakthroughs in Majorana qubit readout, scalable ion-trap systems...

2. Quantinuum Helios System: Most Accurate Quantum Computer
   URL: https://www.perplexity.ai
   Snippet: Quantinuum's Helios system launched November 2025, most accurate yet...

3. Google Quantum Echoes and Topological Superconductors
   URL: https://www.perplexity.ai
   Snippet: Google's Quantum Echoes algorithm demonstrates quantum advantage...

How It Works

1. Web Search Agent — Searches using Perplexity's sonar model, real-time with citations
2. Document Reader — Fetches and parses full content from URLs
3. Summarizer — Uses Groq LLM to synthesize documents into coherent findings
4. Critic — Scores 1-10 based on factual grounding and comprehensiveness
5. Reflection Loop — If score < 7 and iterations < 3, loops back to web search for refinement

Project Structure

research-assistant/
├── agents/                  # Individual agents
│   ├── web_search_agent.py
│   ├── document_reader_agent.py
│   ├── summarizer_agent.py
│   └── critic_agent.py
├── graph/                   # Orchestration
│   ├── state.py             # ResearchState TypedDict
│   └── supervisor.py        # LangGraph routing
├── evals/                   # Testing
│   ├── golden_set.json
│   └── run_evals.py
├── main.py                  # CLI entrypoint
├── requirements.txt
├── .env.example
└── LICENSE

Run Evaluations

python evals/run_evals.py

Tests against 5 golden-set research questions with keyword and citation validation.

Troubleshooting

GROQ_API_KEY not found

• Ensure .env exists in project root
• Check: echo $GROQ_API_KEY
• On Windows: restart terminal after setting .env

Document parsing failed

• Some sites block requests; add more sources
• Timeout is 5 seconds

Score stays below 7

• Try a more specific question
• Check API keys are valid

License

MIT License - see LICENSE for details.

---

Built with LangGraph, Perplexity AI, and Groq