Senior engineer who builds the system that 𝗿𝗲𝗰𝗼𝗿𝗱𝘀 𝗵𝗼𝘄 𝗮 𝗽𝗲𝗿𝘀𝗼𝗻 𝘂𝘀𝗲𝘀 𝗮 𝘄𝗲𝗯 𝗮𝗽𝗽𝗹𝗶𝗰𝗮𝘁𝗶𝗼𝗻 and replays it exactly, so that an 𝗮𝗴𝗲𝗻𝘁 𝗰𝗮𝗻 𝗹𝗲𝗮𝗿𝗻 𝗮 𝘁𝗮𝘀𝗸 𝗳𝗿𝗼𝗺 𝗮 𝗱𝗲𝗺𝗼𝗻𝘀𝘁𝗿𝗮𝘁𝗶𝗼𝗻 and be tested against what really happened. Greenfield.

THE TECHNICAL CHALLENGE

We build agents on open-weight LLMs that operate web applications through the browser. One of the ways an agent learns a new application is from a person's recorded demonstration. The recording holds the page, network traffic, screen images and every click and keystroke on one timeline. A replay runs on the recording alone and marks every point where its result differs. Sensitive data is removed from the recording, and the replay still works. Embedded frames, changing identifiers and expiring sessions make a replay look correct when it is wrong. The recording libraries exist. The engineering is one timeline, aligned clocks and a replay that gives the same result every time. Progress is measured against benchmarked results.

KEY RESPONSIBILITIES

• Build the recording of a browser session end to end: page, network traffic, screen images, clicks and keystrokes, and timing on one timeline, across tabs and embedded frames

• Replay against what was recorded, the same way every time, show every point where the replay differs, and show that it matches the recording

• Remove sensitive data from the recording, following a written rule, without breaking the replay

• Turn recordings into what the agent learns from and is tested against, in versioned formats

DESIRED QUALIFICATIONS

• A recorder, replayer or trace tool built or maintained with external users: rrweb or a company's own version of it, Playwright tracing, screen recording over the Chrome DevTools Protocol, Webrecorder tooling

• Replay against recorded network traffic (WARC, WACZ, HAR) that shows where a replay differs, or a signed archive format

• A dataset or file format authored for what agents that operate a computer or a browser learn from, with timing and the removal of sensitive data stated

• Human demonstrations turned into automation that works on the next case as well

EXPECTED QUALIFICATIONS

• T-shaped: deep in one domain, with working breadth in a neighbouring one

• Structures a large, incompletely specified problem and drives it to a working result independently; knows where each of the recorded channels can mislead and says so before being asked

• Session recording or replay tooling built, with what went wrong on embedded frames, shadow roots, canvas and media told from experience; several channels put on one timeline with the browser's own clock and real time kept apart; uses AI coding tools daily and verifies their output

• Network recording and replay built (HAR, WARC or comparable), handling tokens, timestamps and expiring sessions; removal of sensitive data that covers the page, the network traffic and the screen images

——

HOW WE WORK

• Product engineering: we own what we build and run it in production

• Small teams, two-week cycles, working software at every review

• AI coding tools are part of the standard workflow

WHAT WE OFFER

• Founding-team scope

• AI-augmented engineering environment

• Access to on-premise Nvidia B200s

• Flexible work environment

PROCESS

• Introductory call

• Technical conversation

• Practical session; the format is agreed with you

In coding exercises, AI tools are allowed and expected. No LeetCode.

WHO WE ARE

New product organisation as part of a large semi-government in Abu Dhabi. International, ex-FAANG team. Completely greenfield, with a modern tech stack.

——

REQUIREMENTS TO BE CONSIDERED

• Clear written and spoken English

• 5+ years building production software; strong TypeScript; own code operated in production, with users

• Deep working knowledge of the Chrome DevTools Protocol and of how Playwright or Puppeteer work internally; a recording, replay or trace system of the candidate's own building, that can be shown

• Bachelor's degree in any field, or self-taught with a track record of open-source contributions

RELATED TECHNOLOGIES AND CONCEPTS

• Browser internals and automation: Chrome DevTools Protocol, Playwright tracing and trace viewer, Puppeteer, rrweb 2.x and its forks, Steel Browser and hosted browsers

• Record and replay: HAR, WARC and WACZ, warcio, Browsertrix, replay from recorded responses, time-travel debugging

• Rendering and input: DOM mutation observers, shadow DOM, canvas and video capture, screencast frames, input dispatch, WebRTC

• Data handling: removal of sensitive data across page, network and screen images, versioned formats for what agents learn from, test infrastructure that gives the same result every time