Sphere: the app, backend and recommendation engine behind a private UK network

The client

Sphere is a private network for finance and technology professionals in the UK. Membership is by approval only, and Sphere's staff vet every applicant. Events sit at the centre of the product, because Sphere believes good connections happen in person.

What they needed

A private network is only worth joining if it introduces you to the right people. Sphere needed an app that takes a member from applying to their first event, suggests who to meet and explains why. Its staff needed one place to vet members, run events and manage company memberships. Behind both sat one backend that handles payments in pounds with UK VAT and never lets anyone in without approval.

What we built

We built the whole platform: the iOS and Android app, the backend API, the recommendation engine and the staff portal.

The mobile app

• Joining. Apply with an email or through LinkedIn, verify your email, and follow your application while staff review it. Company members verify their work email to get the seat their employer pays for.
• Discover. Suggested members with the reasons to connect, plus search and filters to find more.
• Connections, one to one chat and push notifications.
• Events. Browse what is coming up, see it on a map, ask to attend and share an invite. Hosts invite members, approve requests, message everyone attending, mark attendance at the door and export the guest list.
• Communities. Members join smaller groups inside Sphere and post to them.
• Premium. Upgrade through Stripe Checkout and manage billing in Stripe's customer portal.

The recommendation engine

• A people feed of the twenty most relevant members, each scored from 0 to 100 on signals a person can check, such as sector fit, mutual connections, seniority and distance.
• A few short reasons with every suggestion, written by Claude on Amazon Bedrock from the same facts the score used. If the AI call fails, a template built from those facts takes its place, so members never see an error.
• Weekly introductions: four for premium members and one for free members, never repeating a person within eight weeks.
• Networking goals. A member picks what they are on Sphere for, and the feed follows it.
• A first draft of a member's bio. The member edits and saves it, never the AI.

Before changing any weights, we asked a group of members to rate the people we recommended to them. Each rating is stored next to the score that produced it.

The staff portal

• Review of every new registration, with an AI summary of unfamiliar companies that applicants name.
• Event creation, invitations and Outlook calendar invites sent straight to attendees.
• Company memberships, seats, tiers and billing, plus a dashboard where each company sees its own employees, events and seats.
• Reports and member analytics with Excel export, a recommendations dashboard and an audit log of staff actions.

The backend

One Go API serves the app, the staff portal and the company dashboard. Members subscribe through Stripe Checkout. Companies pay through a payment link and assign seats. After one or two failed payments a member keeps access, after the third it pauses, and a paid invoice restores it straight away. Marketing records in HubSpot update in the background, and background jobs handle calendar invites, match suggestions and clean up.

How it runs

The API is written in Go and runs on AWS Fargate in the London region, with PostgreSQL on Amazon RDS and media in Amazon S3. Database migrations run when the server starts, and AWS CodePipeline deploys to development and production. The recommendation engine is a Python and FastAPI service with Redis, triggered weekly by Amazon EventBridge. The staff portal is Django with a React front end. All of it is defined in Terraform.

The app is one React Native codebase with Expo for both platforms. It checks its own version on launch and sends members to the store when an update is required. Errors report to Sentry.

If your community needs a platform like this

If you run a membership network, from the app in members' hands to the tools your team uses, the same approach works for you. Gate access on approval, keep payments and access in step, and let AI explain each introduction, not decide it. Talk to us about your platform.