LeadPilot
An AI chat agent that lives on a business website, qualifies every visitor like a sales rep, scores the lead and drafts the follow-up email, so no enquiry is left waiting.
The problem
Small businesses lose enquiries because nobody replies in time, and the replies they do send often skip the basics: what the visitor needs, their budget and their timeline.
The approach
A chat widget greets each visitor and asks the right questions one at a time. When the conversation ends, a second AI step extracts the lead details, scores the lead from 0 to 100, writes a short summary with a next step and drafts the follow-up email.
How it is built
A FastAPI backend serves a vanilla JavaScript chat widget that can be added to any site with one script tag. Conversations and leads are stored in SQLite. The LLM client works with any OpenAI-compatible API or Gemini and returns structured JSON for the lead fields and score. Without an API key it falls back to a rule-based demo mode. An optional webhook can send each new lead to another tool.
My role
Full design and development of LeadPilot.
Delivered functionality
A working chat widget, a demo agency website, a lead dashboard with Hot / Warm / Cold filters, AI summaries, follow-up email drafts, status tracking and re-scoring.
What the agent does
- Chat: greets visitors and asks about their need, budget, timeline and contact details.
- Score: rates every lead from 0 to 100 and labels it Hot, Warm or Cold.
- Summarise: writes a plain-English summary and a clear next step for each lead.
- Follow up: drafts a ready-to-send email in the business's own name.
- Dashboard: all leads in one place with the full conversation, status, copy-email and re-score buttons.
- Configurable business profile, so one agent can serve different businesses.
Inside LeadPilot
DEMO VIDEOScreen recording of the working agent on a sample agency website: a visitor chats, the lead appears in the dashboard with a score, an AI summary and a follow-up email.