Turn Your Chatbot Into a Revenue Engine

RAG chatbot engineered to convert website visitors into booked strategy sessions — not just answer FAQs.

Video Demo

Ideal Para

Professional service firms where every booked call is worth $500+
Businesses with strong website content that visitors never find
Teams that want leads qualified 24/7 without hiring SDRs

No Es Para

E-commerce businesses needing product recommendation engines
Companies without clear service offerings or pricing
Teams looking for a generic FAQ deflection tool

Lo Que Obtienes

Revenue-first chatbot with sales CTAs woven into every conversation path
Dual-source RAG pulling from your website content and curated knowledge base
Embeddable widget deployable on any domain with one script tag
Admin dashboard with conversation analytics and knowledge management

El Problema

Language schools treat chatbots as deflection tools — reduce ticket volume, speed up FAQ answers, dead-end the conversation. The result is a cost center that never generates revenue. Meanwhile, anonymous website visitors leave without converting because there's no sales-aware system guiding them toward high-value actions like booking a strategy session.

La Solución

An AI chatbot explicitly engineered as an automated sales agent. Every architectural layer — system prompts, RAG citations, error handling, rate limits — is designed to drive conversational conversion. The bot uses benefit-driven language with real client social proof, links RAG source citations to high-value service pages, and hard-codes CTAs for free strategy sessions directly into the system prompt.

Características Principales

Revenue-first RAG architecture merging website content and manual knowledge sources (top 5 by cosine similarity)
System prompts hard-coded with sales CTAs — every conversation steers toward booking a strategy session
Dual-source knowledge base: scraped website content (0.5 threshold) + curated documents (0.6 threshold)
Source citations that drive traffic to high-value service pages, increasing time-on-site and eliminating hallucination
Full EN/ES bilingual support with locale-aware personality and benefit-driven language
Embeddable widget (iframe + script tag) deployable on any client domain with CORS configuration
Admin dashboard with chat analytics, knowledge management, and bot settings

Resultados

5-layer revenue engine: System Prompts, Error CTAs, Localization, Rate Limits as Strategy, Source Citations

2 RAG knowledge sources merged by cosine similarity with configurable thresholds

Sub-second streaming responses via Vercel AI SDK with OpenAI GPT-4o

Zero-auth guest experience via sessionId cookies — no signup friction before conversion

Architecture Overview

NY AI Chatbot is an enterprise-grade conversational AI platform built as a revenue engine, not a support tool. The core design principle is "Conversational Conversion" — every layer of the system is architected to guide anonymous visitors toward booking a paid strategy session. RAG isn't just for accuracy; source citations deliberately link to high-value service pages. Rate limits aren't just for protection; they create urgency. Error messages aren't just for UX; they embed CTAs.

The system runs on a dual-source RAG pipeline: scraped website content (cosine similarity threshold 0.5) and manually curated knowledge documents (threshold 0.6) are searched independently, then the top 5 results are merged by similarity score and injected into the system prompt with source URLs.

Key Engineering Decisions

Revenue-first prompt architecture over neutral FAQ: The system prompt in lib/ai/prompts.ts explicitly instructs the bot to offer free 30-minute strategy sessions and link to the booking page. This isn't a feature toggle — it's structural. The bot's personality, language, and CTAs are hard-coded to sell, using benefit-driven framing ("advance your career") rather than feature descriptions ("we teach grammar").

Dual RAG sources over single knowledge base: Website content is ingested via automated scraping with text-embedding-3-small embeddings stored in pgvector. Manual document uploads allow the admin to inject curated sales content that doesn't exist on the public website. Both sources are searched in parallel and merged, giving the bot access to both public-facing content and internal sales messaging.

SessionId cookies over mandatory auth: Anonymous visitors get a frictionless chat experience via sessionId cookies from lib/session.ts — no signup wall between the visitor and the conversion funnel. Clerk authentication is reserved exclusively for the admin dashboard.

Embeddable widget over standalone app: The chatbot deploys as a script tag + iframe combination served from /api/embed, allowing the client to drop it onto any page of their existing website. CORS is configured per-domain in lib/security/cors.ts.

The Five Gears

The Brain — System prompts and persuasion logic in lib/ai/prompts.ts
The Trigger — Error messages and rate limit responses repurposed as CTA delivery mechanisms
The Chameleon — Full EN/ES localization with culturally-adapted personality
The Soft Wall — Rate limits positioned as strategic nudges toward booking
The Bridge — RAG source citations linking directly to high-value service pages

Security

Clerk middleware protecting all /admin and /api/admin routes
IP-based and session-based rate limiting on the chat API
Input validation and sanitization via lib/security/validation.ts
CORS allowlist for embed widget domains
No credentials or secrets in client-side code

¿Listo para discutir una solución similar?

Exploremos cómo la automatización con IA puede ayudar a tu negocio.

Agendar una Consulta