Skip to main content

FiftyFive Tech

Email Us sales@fiftyfivetech.io
Get in Touch
Rows of dark server racks with small blue and green lights stretching across a large data centre hall with steel beams and ceiling lights

DevOps & platform engineering

Multi-tenant AI app hosting on Kubernetes at 25% lower compute cost

FiftyFive built a multi-tenant AI hosting platform on Kubernetes with tenant isolation, RAG chatbots, Stripe billing and about 25% lower compute costs.

Discuss running AI workloads
Industry
Enterprise IT — AI application hosting
Region
Middle East
Solution
Multi-tenant AI hosting platform on Kubernetes
Story

About the client

The client builds digital platforms for application deployment and wanted to offer AI hosting as a service. Its target users were business teams without infrastructure skills. Normally, putting an AI app or chatbot online means provisioning compute, configuring domains and certificates, and arranging invoicing by hand. The client wanted all of that hidden behind a simple interface, with compute cost, security and tenant separation kept under control.

Challenge

AI hosting non-technical teams can run themselves

  • Let non-technical users deploy AI services through a guided interface, without touching infrastructure
  • Allocate GPU and CPU resources dynamically while keeping compute costs down
  • Meter usage and invoice automatically, with no manual billing step
  • Isolate each tenant with SSL, its own subdomain and secure ingress
  • Support several roles with project-specific permissions across tenants
  • Monitor GPU and CPU workloads in real time and alert on anomalies.
Solution

What FiftyFive built

A four-person FiftyFive team, a project lead and three DevOps engineers, built the platform to production. The scope covered: a React.js frontend that guides non-technical users through deploying AI services; Node.js and TypeScript backend microservices for orchestration, tenancy and platform logic; Docker and Kubernetes orchestration of AI workloads on AWS and RunPod GPU nodes, with pre-built Ubuntu and LLM container images; a RAG pipeline for knowledge-base chatbots using OpenAI and other LLM APIs, with one-click deployment; hourly, usage-based invoicing through the Stripe Billing API; Google OAuth sign-in, JWT sessions and tenant-level role-based access control; a GitHub-linked CI/CD pipeline and GPU and CPU utilisation dashboards with anomaly alerts.

How the solution works

  1. Tenant onboarding and deployment

    • A user chooses an AI app or chatbot in the React interface and deploys it with one click.
    • Backend microservices create the tenant's workload from a pre-configured Ubuntu and LLM container image, which shortens onboarding because nothing is built from scratch.
  2. Compute orchestration

    • Kubernetes schedules containerised workloads across AWS and RunPod GPU nodes as well as CPU capacity.
    • Placing GPU work across the two providers is the basis of the compute-cost reduction.
  3. Tenant isolation

    • Kubernetes Ingress gives each tenant its own subdomain and SSL certificate and keeps its traffic isolated.
    • Google OAuth with JWT sessions authenticates users, and role-based access control applies project-specific permissions within each tenant.
  4. AI services

    • For knowledge-base chatbots, a RAG pipeline retrieves the tenant's content and passes it to OpenAI or other LLM APIs to generate answers.
  5. Billing and operations

    • Usage is metered per tenant and invoiced hourly through the Stripe Billing API.
    • A GitHub-linked CI/CD pipeline deploys platform changes, and dashboards show GPU and CPU utilisation with real-time anomaly alerts.

Technology stack

React.jsGuided deployment interface for non-technical users
Node.js, TypeScriptBackend microservices for orchestration and tenancy
Docker, KubernetesContainerised AI workloads and scheduling
AWS, RunPod GPU nodesGPU and CPU compute
Kubernetes IngressPer-tenant subdomains, SSL and traffic isolation
OpenAI / LLM APIs with RAGKnowledge-base chatbots
Stripe Billing APIHourly usage-based invoicing
Google OAuth, JWT, RBACAuthentication, sessions and tenant permissions
GitHub-linked CI/CDOne-click deployment workflow

What the solution enabled

Outcomes

Business transformation

About 25%

Lower compute cost

In minutes

AI apps deployed by non-technical teams

Hourly

Usage-based Stripe billing replaced manual invoicing

About 25% lower compute cost

Placing GPU workloads across AWS and RunPod nodes cut compute costs by roughly a quarter.

AI apps deployed in minutes

Non-technical teams launch AI apps and chatbots in minutes instead of waiting for manual provisioning.

Automated invoicing

Hourly, usage-based Stripe billing replaced manual invoicing.

Isolated tenants

Each tenant runs on its own subdomain, certificate and ingress, with project-level permissions.

FiftyFive Technologies built a production multi-tenant platform for a Middle East enterprise IT company that wanted non-technical business teams to deploy AI apps and chatbots without provisioning servers or handling invoices. The platform runs containerised AI workloads on Kubernetes across AWS and RunPod GPU nodes, isolates each tenant with its own subdomain, SSL and ingress, offers retrieval-augmented generation (RAG) chatbots over LLM APIs, and bills usage hourly through Stripe. GPU placement across the two providers cut compute costs by about 25%.

FAQ

Technical questions

How do you isolate tenants on a shared Kubernetes AI hosting platform?

In this platform, FiftyFive Technologies configured Kubernetes Ingress so each tenant has its own subdomain, SSL certificate and isolated traffic path. Google OAuth with JWT sessions authenticates users, and role-based access control applies project-specific permissions inside each tenant.

How can GPU hosting costs be reduced for AI workloads?

FiftyFive Technologies scheduled containerised AI workloads with Kubernetes across both AWS and RunPod GPU nodes, placing work on cost-competitive capacity. For a Middle East enterprise IT company, this approach reduced compute costs by about 25%.

Talk to FiftyFive's DevOps engineers about running AI workloads on cost-efficient infrastructure.

Book a call

Project snapshot

Industry
Enterprise IT — AI application hosting
Region
Middle East
Services
Product development, frontend and backend development, DevOps, integration, AI integration
Team
4: project lead and three DevOps engineers
Duration
Since May 2025 (ongoing)
Status
In production
Core technologies
React.js, Node.js, TypeScript, Docker, Kubernetes, AWS, RunPod, Stripe