How to Build an Autonomous B2B Lead Generation Engine with Playwright, Groq & Resend

The Project

NexusScout AI is an autonomous B2B lead generation engine powered by a three-agent AI pipeline. It scrapes Google Maps for businesses, enriches their data by extracting emails from their websites, and sends personalized outreach emails — all from a single web UI.

The problem it solves: manual lead generation is slow, repetitive, and scales poorly. Sales teams spend hours searching Google Maps, visiting websites one by one, copying emails, and drafting outreach. NexusScout automates the entire pipeline end-to-end.

Architecture

User (React UI)
    │
    ▼
FastAPI Backend
    │
    ├── /api/scout       → Agent 1: Scouter (Playwright + Stealth)
    ├── /api/leads       → Agent 2: Enricher (Email extraction)
    ├── /api/leads/contact → Agent 3: Outreach (Groq + Resend)
    │
    └── SQLite Database (leads.db)

Three-Agent Pipeline

Agent 1: Scouter

The Scouter uses Playwright with stealth plugins to search Google Maps without getting blocked. It takes a query like "Web Design Agencies in London" and extracts:

• Business name
• Website URL
• Address
• Rating
• Category

# Simplified scouter logic
from playwright.sync_api import sync_playwright
from playwright_stealth import stealth_sync

def scout(query: str) -> list[dict]:
    with sync_playwright() as p:
        browser = p.chromium.launch(headless=True)
        page = browser.new_page()
        stealth_sync(page)
        
        page.goto(f"https://www.google.com/maps/search/{query}")
        page.wait_for_selector('a[href*="/maps/place"]')
        
        results = []
        for el in page.query_selector_all('a[href*="/maps/place"]')[:20]:
            name = el.get_attribute("aria-label")
            href = el.get_attribute("href")
            results.append({"name": name, "maps_url": href})
        
        return results

Agent 2: Enricher

Once the Scouter returns businesses with websites, the Enricher visits each site and scrapes contact information. It handles:

• Email extraction from contact pages, about pages, and footers
• Social media profile discovery
• Business categorization based on site content

async def enrich_lead(website: str) -> dict:
    emails = await extract_emails(website)
    social = await find_social_profiles(website)
    category = await categorize_business(website)
    
    return {
        "emails": emails,
        "social": social,
        "category": category,
        "enriched": True
    }

Agent 3: Outreach

The Outreach agent is where the AI magic happens. It uses Groq's Llama 3.3 70B to generate personalized email opening lines based on the lead's business type and website content.

from groq import Groq

def generate_outreach(business_name: str, website: str, category: str) -> str:
    client = Groq(api_key=settings.GROQ_API_KEY)
    
    prompt = f"""Write a short, personalized email opening line 
    for {business_name} (a {category} business at {website}).
    Keep it professional but warm. Max 2 sentences."""
    
    response = client.chat.completions.create(
        model="llama-3.3-70b-versatile",
        messages=[{"role": "user", "content": prompt}],
        temperature=0.7,
    )
    
    return response.choices[0].message.content

Then Resend delivers the email:

import resend

resend.Emails.send({
    "from": "onboarding@resend.dev",
    "to": lead.email,
    "subject": f"Quick question for {lead.business_name}",
    "html": f"<p>{ai_generated_message}</p>"
})

Tech Stack

UI Features

• Stats Bar — Real-time counts (Total / Enriched / Failed)
• Search Bar — Type a query → Scouter scrapes Google Maps
• Lead Cards — Expandable details, email, website, status badges
• Contact Button — Triggers AI personalization + email delivery
• Delete — Remove leads with one click

Quick Start

# Backend
cd backend
uv add .
uv run uvicorn main:app --reload

# Frontend (separate terminal)
cd frontend
npm install
npm run dev

Required API Keys

GROQ_API_KEY=gsk_your_key_here
RESEND_API_KEY=re_your_key_here

What I Learned

1. Three-agent architecture beats monolithic scrapers — Separating search, enrichment, and outreach into independent agents means each can fail and retry without blocking the others. If the Enricher can't find an email, the Scouter's results aren't lost.

2. Playwright stealth is essential — Google Maps blocks headless browsers aggressively. The playwright-stealth library patches detected automation signals and dramatically improves success rates.

3. AI personalization at scale — Writing personalized emails for 100+ leads manually is impractical. Groq's Llama 3.3 generates convincing opening lines that reference each business's specific category and website — far better than generic templates.

4. Async SQLAlchemy makes a difference — When the Enricher is processing 50+ websites concurrently, async database operations prevent write contention and keep the UI responsive.

5. Resend is production-ready for transactional email — Simple API, good deliverability, and generous free tier. Perfect for an MVP lead generation tool.