PROJECT TITLE
LeadBot Partners - AI Chatbot Builder
My Role
I led the end-to-end design, development, and deployment of the LeadBot Partners platform — an AI-powered chatbot builder SaaS for lead generation.
My responsibilities included product design, full-stack engineering, and AI architecture. I built the chatbot engine with RAG-powered knowledge bases (Pinecone + AWS Bedrock), lead capture forms with conditional logic, embeddable chat widgets, and Calendly booking integration.
I also designed and implemented the Voice AI Receptionist feature — a real-time phone-based AI system using Twilio WebSocket streaming, Deepgram speech-to-text, Claude for conversation intelligence, and ElevenLabs for natural text-to-speech. Additionally, I integrated Stripe for tiered subscription billing and built the analytics dashboard.
What is LeadBot Partners
LeadBot Partners is an AI-powered chatbot builder platform designed specifically for lead generation and customer engagement in professional services.
The platform enables law firms, business coaches, therapists, real estate agents, and financial advisors to deploy intelligent chatbots that qualify leads 24/7, capture contact information, and book consultations — all without technical expertise.
Key differentiators include a Voice AI Receptionist that answers phone calls with natural conversation, a RAG-powered knowledge base supporting PDFs, URLs, YouTube videos, and FAQs, and a fully embeddable widget system with 7+ platform integrations — delivering a complete lead automation solution with Stripe-powered tiered billing.
1. The Problem
90% of website visitors leave without contacting businesses. Professional service providers miss leads during off-hours, lose prospects to abandoned intake forms, and waste time on unqualified inquiries. <br/><br/> Existing chatbot solutions lack the intelligence, customization, and voice capabilities needed for high-value professional services.
a. User Challenges: Professional service providers — especially solo practitioners and small firms — struggle with missed after-hours calls, abandoned intake forms, and the inability to qualify leads before consultations. They need 24/7 intelligent engagement without hiring additional staff.
b. Business Challenges: Building a multi-tenant SaaS platform that delivers real-time AI chat and voice interactions, handles knowledge base processing at scale, supports embeddable widgets across diverse platforms, and manages tiered subscription billing with usage-based limits.
Law Firm Owner Persona
Michael, a solo immigration attorney, misses 40% of potential client calls after hours. He needs an AI receptionist that can answer the phone, qualify callers based on case type, capture their information, and book consultations — all without hiring a receptionist.
Business Coach Persona
Lisa, a business coach running online programs, wants a chatbot on her website that answers FAQs from her course content, qualifies potential coaching clients, and books discovery calls through Calendly — automatically converting visitors into booked leads.
Frontend
Built with Next.js 16 and TypeScript, featuring Tailwind CSS 4, Radix UI components, Framer Motion animations, GSAP scroll reveals, and Recharts for analytics visualization.
Backend
Next.js API routes with server components, plus a separate Express.js WebSocket server for real-time voice AI processing.
Database
PostgreSQL via Supabase with Prisma ORM for multi-tenant data — chatbots, conversations, leads, knowledge bases, voice calls, and subscription management.
AI Integration
AWS Bedrock (Claude 3.5 Haiku) for chat intelligence, Amazon Titan for embeddings, and Pinecone vector database with per-chatbot namespace isolation for RAG.
Voice AI Stack
Twilio for phone numbers and WebSocket audio streaming, Deepgram Nova-3 for real-time speech-to-text, ElevenLabs Flash v2.5 for natural text-to-speech, with mu-law audio conversion.
Billing
Stripe integration for 4-tier subscription management (Free, Basic, Pro, Agency) with webhook-based updates, customer portal, and usage-based feature gating.
Infrastructure
AWS S3 for file storage, AWS Lambda for async YouTube transcript processing, Redis/Valkey for conversation caching, and Resend for transactional emails.
a. RAG-Powered Chatbot Engine
- Built a multi-tenant chatbot engine with AWS Bedrock (Claude 3.5 Haiku) and Pinecone vector database, supporting per-chatbot namespace isolation for knowledge base retrieval.
- Implemented multi-format knowledge ingestion — PDFs, URLs, YouTube transcripts, and FAQs — with real-time processing status tracking.
b. Voice AI Receptionist
- Architected a modular real-time voice pipeline: Twilio WebSocket → Deepgram STT → Claude AI → ElevenLabs TTS, with mu-law audio conversion and sentence-level streaming.
- Built business hours scheduling, voicemail fallback, call recording/transcription, lead capture during calls, and call transfer capabilities.
c. Lead Capture & Embedding System
- Designed dynamic lead capture forms with conditional logic, multi-step flows, and automatic email notifications on lead submission.
- Built an embeddable widget system with customizable branding, suggested questions, and public API endpoints for cross-platform integration.
d. Subscription Billing & Analytics
- Integrated Stripe for 4-tier subscription management with webhook-driven plan synchronization, usage-based feature gating, and customer portal access.
- Built an analytics dashboard with area charts for conversation/lead trends, voice call statistics, and period comparison views (7/30/90 days).
- Designed a modular voice provider architecture (swappable STT/TTS/LLM components) achieving 95% cost savings over monolithic alternatives.
- Used Server-Sent Events for streaming chat responses with progressive token disclosure for better perceived performance.
- Implemented Redis/Valkey hot caching for conversation context and RAG results, reducing repeat query latency significantly.
- Built async processing via AWS Lambda for YouTube transcript extraction and knowledge base embedding generation without blocking the UI.
