Conversational AI SaaS

AI Assistant

AI Chatbots That Learn Any Business in Minutes

AI Assistant is a self-serve SaaS platform that turns any business's website and documents into a trained, branded AI chatbot — no code required.

We designed and built it end to end: product design, frontend, backend, the RAG pipeline, infrastructure, and deployment.

Project Overview

AI Assistant is a self-serve SaaS platform for building custom AI chatbots. A business points the platform at its website or uploads PDFs, and the platform reads, indexes, and turns that content into a branded chat widget that answers customer questions accurately.

The widget goes live on any site with a single embed snippet, streaming answers in real time — grounded in the business's own content rather than the model's general knowledge.

Our team owned the product from end to end: product design, frontend, backend, the retrieval-augmented generation pipeline, infrastructure, and deployment.

End-to-End Build

Engagement

Vercel + Render

Infrastructure

Conversational AI

Industry

SaaS + Chat Widget

Platform

Client Challenges

Small and mid-sized businesses answer the same customer questions over and over — pricing, policies, product details, opening hours — and their options were limited:

01

More support staff is expensive and still doesn't cover 24/7

02

Rule-based chatbots need every flow scripted and break on new phrasing

03

Generic AI chatbots sound fluent but hallucinate prices, policies, and features

04

Enterprise conversational AI is too costly and slow to set up without engineers

05

Business content lives across modern, JavaScript-heavy sites and PDFs

06

AI usage costs needed to stay predictable for both customers and the platform

The goal was a product a non-technical owner could set up in one sitting — answering only from their own content, matching their brand, with predictable costs.

Our Approach

We built the platform around one simple workflow — connect your content, customise your assistant, paste one line of code:

  1. Step 1

    Content ingestion: a headless Chromium crawler for websites, plus PDF upload and re-indexing

  2. Step 2

    A RAG pipeline with embeddings in Postgres + pgvector behind an HNSW index

  3. Step 3

    A lightweight embeddable widget streaming replies in real time over WebSockets

  4. Step 4

    Self-serve branding: logos, brand colours, and per-widget behaviour

  5. Step 5

    Atomic token metering for usage-based pricing and cost protection

  6. Step 6

    Consolidating four external services onto Supabase and Resend

Grounding beats generality — customers care less about how clever the AI sounds than whether it's right about their business.

Key Features Delivered

Website Crawling & PDF Ingestion

  • Puppeteer crawler renders JavaScript-heavy sites fully
  • Upload manuals, policies, and price lists as PDFs
  • Re-index content whenever it changes

Retrieval-Augmented Generation

  • Content chunked, embedded, and stored in pgvector
  • HNSW index for fast similarity search
  • Answers grounded in each widget's own knowledge base
  • Sharply fewer hallucinations than a generic chatbot

Embeddable Real-Time Widget

  • Installs on any site with a single snippet
  • Replies stream live over Socket.IO
  • Lightweight and responsive on every device

Branding & Customisation

  • Logo upload and built-in colour picker
  • Per-widget behaviour settings
  • Feels native to the business's own site

Usage-Based Pricing

  • Every conversation metered by tokens
  • Atomic usage counters in the database
  • Usage visible to customers in the dashboard
  • Protection against runaway AI costs

Dashboard & Administration

  • Usage analytics and charts
  • Multiple widgets per account
  • Role-based access for admins and users
  • Admin panel for users, content, and settings

Security & Multi-Tenancy

  • JWT auth with bcrypt-hashed passwords
  • Supabase Row Level Security for tenant isolation
  • Helmet and rate limiting on public endpoints
  • Email confirmation and password reset

Project Gallery

A look at AI Assistant's dashboard, widget customisation, content ingestion, and the live chat widget embedded on a customer's site.

Project Image 1
Project Image 2
Project Image 3
Project Image 4
Project Image 5

Technology Stack

Frontend

Next.jsReactMaterial UI

Backend & Real-time

Node.jsExpressSocket.IO

AI & Retrieval

OpenAILangChainpgvector

Data & Storage

SupabasePostgreSQLSupabase Storage

Crawling & Email

PuppeteerResend

Hosting

VercelRender

Our Main Contributions

Product design for a one-sitting, self-serve onboarding flow

Next.js marketing site and responsive customer dashboard

Node.js / Express APIs for auth, widgets, users, and support

Puppeteer crawling and PDF ingestion pipeline

RAG pipeline with OpenAI, LangChain, and pgvector

Real-time embeddable chat widget over Socket.IO

Token metering, usage analytics, and role-based admin panel

Migration from MongoDB, AstraDB, S3, and SendGrid to Supabase and Resend

Impact & Results

Businesses go from a website URL to a live, trained chatbot in a single sitting

Routine customer questions answered instantly, 24/7, without a human

RAG-grounded answers drawn from each business's own content, with far fewer hallucinations

External managed services halved — four providers consolidated to two

Application data and vector embeddings in one database, keeping search and metadata consistent

Predictable AI costs through atomic, per-conversation token metering

AI Assistant shows what production AI looks like for small businesses: self-serve setup, answers grounded in real content, and infrastructure lean enough to run profitably.

Ready to Launch Your Project Today?

Kickstart your vision with our expert development team.

Get Started