LEAD AGENT
From Google Maps to a scored outreach list.
Toolkit is built and runs end-to-end. It produces scored leads for outreach — no outreach campaign is running.

Overview
An automated lead-generation pipeline: it discovers businesses in a city and niche, finds their websites and emails, scores each lead with an AI agent under strict rules, writes a personalized opener, and exports everything to a color-coded Excel sheet.
This is the actual lead engine behind Aethelion's outreach plans — built as working code, not a mockup.
Problem
Manual prospecting is slow and unscalable, and 'AI-generated leads' tools often fabricate contacts. The pipeline enforces a strict scoring rubric and only surfaces real businesses with real sources.
Approach
- Google Maps discovery via SERPAPI, paged until the target count is reached.
- Email enrichment through Hunter.io domain search.
- A CrewAI scoring agent applies an exact, deterministic scoring rubric (reviews, rating, website, phone, email) and returns SCORE / REASON / OPENER.
- Leads sorted by score and color-coded (green high, yellow medium, red low) in an .xlsx export.
Architecture
INTENT: this is the actual design the system was built against — not a decorative diagram.
Implementation
- Single Python script orchestrating the whole pipeline (verified on disk, ~260 lines).
- Pagination loop for discovery; graceful empty-result guard.
- Deterministic scoring rubric embedded in the agent backstory; fallback to score 5 on parse failure.
- Real exports on file: Dubai real-estate leads, London real-estate leads, New York web-design-agency leads.
Technology
Workflow
- Enter city and business niche.
- Discover businesses from Google Maps.
- Enrich each with an email via domain search.
- AI agent scores each lead against the strict rubric.
- Export sorted, color-coded leads to Excel.
My Role
- Wrote the entire pipeline, defined the scoring rubric, integrated CrewAI/Groq/SERPAPI/Hunter.io, and produced the export logic.
Demo
Real outputs
Three completed lead sheets (Dubai, London, New York) exist as real exports from this pipeline.
Source
- PycharmProjects/PythonProject/lead_agent.py — full source.
- Exports: Dubai_real_estate__leads.xlsx, london_real_estate_leads.xlsx, New_York_Web_Design_Agencies_leads.xlsx.
Status
Verified
- Full source code exists and is coherent (imports, API calls, CrewAI agent, exporter).
- Three real xlsx outputs exist on disk.
- Scoring rubric is explicit and testable.
- CrewAI execution proven with the companion test script.
Planned / Out of scope
- Automated outreach from the exported list.
- Moving the pipeline into a service with a scheduler.
Separating verified from planned is a design principle of this portfolio — not a footnote.
No live URL is shown because this project is not publicly deployed. If that changes, the link appears here automatically.