geneva (Bumble)

· organisations

founding platform engineer · new york · june 2019 to november 2022.

impact

  • a social messaging app taken completely from conception to over 100,000 daily active users and a $12 million series a.
  • acquired by bumble inc. (nasdaq: bmbl) in 2024 for its technology platform. in 2025 bumble rebuilt its entire backend on it and moved its own members across.
  • for the people using it: messages in 10 ms, real-time updates load-tested to 3 million concurrent users, and a data platform that lifted retention by 30%.
  • for the company: engineering that grew from 7 people to 50-plus across time zones. a feature shipped 1 to 2 weeks faster and deployed a third faster.

the organisation

geneva was the group and community app out of new york: group chats, events and clubs for finding your offline people around a shared interest. coatue backed it. instagram’s co-founders, michael moritz and patreon’s jack conte were among its investors. it raised around $36 million. on 20 may 2024 bumble announced it would acquire geneva to take its friendship product from one-to-one to groups. its chief executive, lidiane jones, said the aim was to “accelerate our friendship product using geneva’s powerful technology platform.” the deal closed that summer. in september 2025 bumble for friends relaunched as a new app built on geneva.

what i did

i owned the go and data function.

  • messages arrived in 10 ms, on small compute. one small instance handled 60,000 messages a minute, so the app stayed fast as it grew.
  • everyone in a group saw a message the moment it was sent. redis was the pub-sub back-plane, load-tested to 3 million people connected at once.
  • nobody missed what mattered to them. notifications fanned out an order of magnitude beyond message volume, from a distributed worker system at 60,000 jobs a second.
  • the app never slowed under its own popularity. a ring buffer cached 500 mb of database objects per instance, and the thundering-herd mitigations held when a whole community opened the app at once.
  • 10-plus features people used every day, from message search to video calls, each one event-driven.
  • what people trusted us with stayed safe, and we could prove it. the security controls, a live log pipeline, and behaviour-driven end-to-end tests of the product as people used it.
  • the long tail hunted down. a 2-day outage traced to a panic inside a vendor’s error-reporting client, written up and fixed in the shared library. a cancellation bug that had made tests flaky for a year, found and removed. structured logging and tracing through every service, so the next one took hours and not days.

what it meant

i raised geneva completely from conception to series b. i learned everything there is to know about taking a startup from an idea to a series b: the tech, the people, the legal. more than that, i learned that i can fight to the fucking death for a product i believe in, and that when you have momentum and heart like that, nothing can stop the company from succeeding.

i put my whole soul into geneva. we believed in it so much that we worked for free inside of it. i was up to 4,000 merge-squashed commits to main a year, each one a different attribute of a feature. i fought to the fucking death to drive this company to success.

the conventions were ours. a startup wrote its own handbook and held itself to the level of the best engineering teams in the world, with one aim: the least cumulative time spent by everyone. the basics of it are what i still do and deliver, everywhere:

  • think for an hour to save a day of coding. the best code is the code not written. a change is composed before i sit down to type it.
  • write to be read once. a title that says what the change is. two sentences on what it is meant to accomplish. nothing i could have said with a thumbs up. agreement is silent; disagreement is a conversation.
  • review every line, and expect the same back. are there tests, are they clear, is there code we do not need, is there a simpler way. an hour in review saves 10 later.
  • let tests drive the design. small functions, pure where they can be. gather the data, then crunch it. unused code is removed, never kept for the day it might be needed.
  • one thing at a time, to completion. say no to the rest, out loud.
  • a pull request goes through one round of review. the self-review happens on the diff before anyone else is asked for their time.
  • a product decision is not a code decision. when something is arbitrary, it is the product’s call, and we ask.
  • understand the technology before using it. every footgun we found had been slapped in without reading the manual.
  • the team’s common spaces, well groomed, used, and digestible by everyone in the company. or removed, and digested by nobody at all.

the rest does not fit in a list. a platform a nasdaq-listed company bought and rebuilt its entire backend on. an engineering team that went from 7 to 50 and shipped faster at 50 than at 7. a codebase where you could not tell who had written what, because the style was the company’s and not the author’s. the hiring, the contracts, the paperwork. the investors’ questions and the numbers that answered them. all of it at once, and it worked.

sources