The problem

SkillBridge Academy needed a way to capture leads, qualify them, and route them to the right sales agent without a person manually reading every form submission and deciding who should follow up.

Why a basic form-to-inbox setup wasn't enough

A form that just emails a team inbox doesn't qualify anyone or route anything, it just moves the manual work downstream instead of removing it. Leads need to be scored, tagged, and handed to the agent best suited to close them, and prospects have questions that shouldn't have to wait for a human to be online.

What I built

An end-to-end lead intake and qualification system built around an n8n form, a Telegram bot named Skye that handles prospect questions using a RAG knowledge base, and automatic routing to the right sales agent based on lead value. High-value leads route to the senior sales agent, nurture leads and general escalations route to the growth sales agent, and every lead gets tracked in a live Google Sheet.

Architecture

Prospect submits n8n form → lead data captured and scored → routed to Amaka Cherish for high-value leads or Abanobi Joseph for nurture leads and escalations → branded confirmation email sent → lead logged in Google Sheets → Skye, the Telegram bot, answers ongoing prospect questions by retrieving answers from a Supabase vector store instead of guessing.

Tools used

n8n, Telegram Bot API, Supabase (RAG vector store), Google Sheets

Proof of concept

Full documentation set built out for the team: onboarding doc, client-facing write-up, demo questions doc, a Word-based RAG knowledge base, three iterations of Skye's system prompt, and branded HTML confirmation emails. Also scoped a Gamma slide deck and brand asset prompts so the system has a consistent look end to end, not just a working backend.

Challenges

Getting Skye to answer prospect questions accurately without hallucinating took multiple passes on both the knowledge base structure and the system prompt. Making sure high-value leads actually routed to the right agent, instead of defaulting to whoever was free, meant building real scoring logic instead of a simple round robin.

What I'd improve next

Add a feedback loop so agent close rates feed back into how future leads get scored and routed, instead of routing logic staying static.