The overview
An AI chatbot platform built for the education sector. The system lets training providers deploy intelligent chatbots that guide learners through course discovery, check funding eligibility, and hand off to human agents via CRM when needed. Over 6,600 live conversations processed and 6,500+ CRM prospects captured.
01
The Problem
Training providers needed 24/7 engagement with prospective learners without overwhelming sales teams. Prospects would visit, browse courses, and leave without converting. No system existed to match learners to courses, verify funding eligibility, or capture intent before it disappeared.
02
System Scale
37 Database Tables
Full relational schema with Row Level Security on every table
45 Edge Functions
Serverless backend handling AI, CRM, messaging, and admin operations
31 Frontend Routes
Complete multi-tenant admin, agent dashboard, and widget interfaces
63+ React Components
Production-grade component library with full TypeScript coverage
03
Conversational AI
A dual-layer NLP system combines rule-based intent classification with an LLM fallback for ambiguous queries. It handles 7 core intent types across 9 conversation flow modes, backed by 30+ specialised prompt templates. A hybrid RAG architecture uses pgvector for semantic search across courses, FAQs, and support content.
04
Course Matching
A multi-stage pipeline extracts topics, runs vector search, then uses AI-powered ranking with category-weighted scoring. The AI Course Picker matches against the full catalogue with cascading fallbacks and deduplication, ensuring learners always see relevant, unique results.
05
Eligibility Verification
Multi-step consent-and-verification workflow with UK postcode-based funding checks. Hard rules like age and visa status bypass AI entirely for instant decisions. The system tracks declines and prevents repeat attempts while keeping the experience smooth.
06
CRM & Handoff
A rules-based handoff engine supports keyword, intent, message count, and confidence-based triggers. Handoffs route agent-to-agent or agent-to-human with a confirmation flow. The CRM layer automates prospect creation with LLM-based data extraction, UK phone/postcode validation, and contact deduplication.
07
Platform Integrations
HubSpot CRM
Full OAuth 2.0 flow with 10 dedicated edge functions for contact sync, custom channels, and bidirectional messaging
WhatsApp Business
Meta Cloud API webhooks with 24-hour session enforcement, template fallback, and rate-limited message splitting
Embeddable Widget
Single script tag deployment with cross-origin messaging, session persistence, and 20+ theming options
Knowledge Management
Document, URL, and text ingestion with automated scraping, paragraph-aware chunking, and separate embedding pipelines
08
The Outcome
The platform is a fully operational enterprise SaaS platform with multi-tenant architecture, role-based access, audit logging, and real-time operations. It has processed over 6,600 live conversations, captured 6,500+ CRM prospects, and serves as the primary sales channel for education providers.
09
Part 9
“The platform completely transformed how we engage with prospective learners. The AI handles the initial conversation intelligently, qualifies leads, and only hands off to our team when it genuinely needs a human touch. Our conversion rates have improved significantly.”
“The platform completely transformed how we engage with prospective learners. The AI handles the initial conversation intelligently, qualifies leads, and only hands off to our team when it genuinely needs a human touch. Our conversion rates have improved significantly.”
The strategic takeaway
What this work demonstrates
The value is not the volume of activity. It is the quality of the decisions the work makes possible.
