Skip to content
All case studies
Case study · Aether AI Support

Building Aether
AI Support.

How we built a modern, AI-powered customer support platform from the ground up using cutting-edge technologies and cloud-native architecture.

6
Months development
5
Firestore collections
120+
API endpoints
2
Client SDKs · TS + Python
Client
Aether AI Support
Industry
Customer Support · AI
Services
Full-stack development
The challenge

Traditional support is broken.

Traditional customer support software suffers from critical limitations. We set out to build a platform that leverages AI to solve these problems while remaining affordable and easy to implement.

  • Slow response times leading to customer frustration and churn
  • Repetitive queries consuming valuable agent time
  • Lack of intelligent routing causing mishandled tickets
  • Disconnected systems creating data silos
  • Complex pricing making it inaccessible for SMBs
Our solution

AI-first, real-time, multi-tenant.

Aether AI Support combines the latest in AI technology with modern web development practices to deliver a transformative support experience.

01

AI-First Architecture

We built a Retrieval-Augmented Generation (RAG) pipeline using OpenAI GPT-4o-mini and ChromaDB vector embeddings. The AI understands context, retrieves relevant knowledge base articles, and generates accurate responses in real-time.

02

Real-Time Collaboration

Using WebSockets and Server-Sent Events (SSE), we implemented instant updates across the entire platform. Agents see live ticket changes, typing indicators, and presence status without page refreshes.

03

Multi-Tenant SaaS Design

The platform supports unlimited companies with complete data isolation. Each tenant can customize their widget, knowledge base, SLA rules, and integrations independently.

04

Serverless Infrastructure

Deploying the backend on Cloud Run and the frontend on Vercel means the platform scales itself with demand and there is no server for anyone to patch, resize or be paged about. A support desk is busiest exactly when its customer is having a bad day, which is the worst possible moment to be provisioning capacity by hand.

Technical deep dive

How it's built.

Backend architecture

FastAPI for its native async support, which is what actually matters here: a support platform spends most of its time waiting on something else, whether that is the model, the vector store or the customer’s own systems. The API is fully typed with Pydantic models, providing automatic validation and OpenAPI documentation.

  • Service-oriented design with dependency injection
  • Background task processing with asyncio
  • Structured logging with correlation IDs
  • Rate limiting and quota enforcement

AI & machine learning

The RAG pipeline processes incoming queries through multiple stages.

  • Query analysis: intent detection and entity extraction
  • Retrieval: hybrid semantic + keyword search
  • Context assembly: relevance scoring and window optimization
  • Generation: GPT-4o-mini with custom system prompts per company
  • Post-processing: confidence scoring and escalation detection

Frontend engineering

Next.js 14 with App Router provides server-side rendering, streaming responses, and optimal bundle splitting.

  • Context-based theming system (light/dark/gradients)
  • Framer Motion animations for smooth UX
  • Cmd+K command palette with keyboard shortcuts
  • Optimistic UI updates with SWR caching
  • Responsive design with Tailwind CSS

Database design

Google Firestore provides automatic scaling and real-time sync capabilities.

  • Companies: multi-tenant configuration and settings
  • Tickets: support requests with full history
  • Messages: conversation threads with metadata
  • Knowledge Base: articles, FAQs, and embeddings
  • Users: agents, admins, and customers
What it changes

What the platform actually does.

RAG
Answers from your own docs
Replies are retrieved from each tenant knowledge base rather than improvised
24/7
Immediate first reply
The assistant responds at once, at any hour, before a human picks it up
Live
Every seat in sync
New tickets and replies appear for agents without anyone refreshing
2
Drop-in SDKs
TypeScript and Python clients, so a product can embed support quickly
Lessons learned

What we'd tell our past selves.

Start with the AI, not the UI

We built the RAG pipeline and AI capabilities first, then designed the interface around them. This ensured the AI was a core feature, not an afterthought.

Invest in real-time from day one

Retrofitting WebSocket support is painful. By designing for real-time updates from the start, we avoided technical debt and delivered a more responsive experience.

Multi-tenancy requires early planning

Every database query, API route, and frontend component needed tenant awareness. Planning this architecture upfront saved months of refactoring.

Type everything

TypeScript on the frontend and Pydantic on the backend caught countless bugs before they reached production. Strong typing is non-negotiable.

Built with

Technology stack.

AI / MLFastAPINext.jsPythonTypeScriptGCPFirestoreOpenAI GPT-4o-miniChromaDBWebSocketsVercelCloud Run
Try it live

Want to try Aether AI Support?

Start your 14-day free trial and experience AI-powered support. Or contact us to discuss building a similar platform for your business.

Visit Aether AI Support Start a project