人工智能研究工程师
AI Research Engineer
关于PostHog
产品开发过去意味着手动编写代码、运行分析、诊断错误，并使用数十种工具推出更改。
PostHog是唯一能作为你的协作者（和你的AI代理）来完成所有工作的平台——自主完成。
我们从开源产品分析开始，从Y Combinator的W20批次中推出 https://posthog.com/handbook/story。此后我们推出了十多个产品 https://posthog.com/products，包括：
- PostHog Code https://posthog.com/code，唯一一个能理解你的产品而非仅仅代码库的AI开发工具。
- 内置的数据仓库 https://posthog.com/docs/data-warehouse，让用户可以使用自定义SQL洞察查询产品和客户数据。
- PostHog AI https://posthog.com/ai，一个由AI驱动的分析师，可以回答产品问题，帮助用户找到有用会话记录，并编写自定义SQL查询。
我们是：
1. 以产品为导向。超过45万个组织已安装PostHog，主要由口碑驱动。我们有非常强的产品市场契合度。
2. 默认活跃 https://paulgraham.com/aord.html。收入增长非常迅速，我们效率很高。我们融资是为了推动雄心和更快增长，而不是维持运营。
3. 获得充足资金。我们从世界顶级投资者处筹集了超过1.8亿美元资金。我们为长期、雄心勃勃的旅程做好了准备。
我们专注于为终端用户打造出色的产品，招聘优秀的团队成员，快速交付，并尽可能地保持独特性 https://posthog.com/deskhog。
我们关注的事情
- 透明：每个人都可以在我们的公开公司手册 https://posthog.com/handbook 中阅读我们的路线图、如何支付（甚至解雇）人员、我们的战略以及我们的工作方式。内部，我们分享收入、董事会会议的笔记和幻灯片，以及融资计划，让每个人都能获得做出良好决策所需的信息。
- 自主：我们不会告诉任何人该做什么。每个人根据对客户影响最大的事情，以及他们觉得有趣和有动力去做的事情来选择接下来要做什么。工程师领导产品团队 https://posthog.com/handbook/wide-company 并做出产品决策 https://posthog.com/handbook/which-products。当需要时，团队灵活且易于改变。
- 快速交付：为什么现在不行？https://posthog.com/handbook/values#why-not-now 我们想打造很多产品；我们不能以正常速度交付。
查看英文原文
ABOUT POSTHOG
Product development used to mean manually writing code, running analysis, diagnosing bugs, and rolling out changes using dozens of tools.
PostHog is the only platform that acts like a co-pilot for you (and your AI agents) to do it all – autonomously.
We started with open-source product analytics, launched out of Y Combinator's W20 cohort https://posthog.com/handbook/story. We've since shipped more than a dozen products https://posthog.com/products, including:
- PostHog Code https://posthog.com/code, the only AI devtool that understands your product, not just your codebase.
- A built-in data warehouse https://posthog.com/docs/data-warehouse, so users can query product and customer data together using custom SQL insights.
- PostHog AI https://posthog.com/ai, an AI-powered analyst that answers product questions, helps users find useful session recordings, and writes custom SQL queries.
We are:
1. Product-led. More than 450,000 organizations have installed PostHog, mostly driven by word-of-mouth. We have intensely strong product-market fit.
2. Default alive https://paulgraham.com/aord.html. Revenue is growing incredibly quickly, and we're very efficient. We raise money to push ambition and grow faster, not to keep the lights on.
3. Well-funded. We've raised more than $180m from some of the world's top investors. We're set up for a long, ambitious journey.
We're focused on building an awesome product for end users, hiring exceptional teammates, shipping fast, and being as weird as possible https://posthog.com/deskhog.
THINGS WE CARE ABOUT
- Transparency: Everyone can read about our roadmap, how we pay (or even let go of) people, our strategy, and how we work, in our public company handbook https://posthog.com/handbook. Internally, we share revenue, notes and slides from board meetings, and fundraising plans, so everyone has the context they need to make good decisions.
- Autonomy: We don’t tell anyone what to do. Everyone chooses what to work on next based on what's going to have the biggest impact on our customers, and what they find interesting and motivating to work on. Engineers lead product teams https://posthog.com/handbook/wide-company and make product decisions https://posthog.com/handbook/which-products. Teams are flexible and easy to change when needed.
- Shipping fast: Why not now? https://posthog.com/handbook/values#why-not-now We want to build a lot of products; we can't do that shipping at a normal pace. We've built the company around small teams – autonomous, highly-efficient groups of cracked engineers https://posthog.com/founders/cracked-manifesto who can outship much larger companies because they own their products end-to-end.
- Time for building: Nothing gets shipped in a meeting. We're a natively remote company. We default to async communication – PRs > Issues > Slack. Tuesdays and Thursdays are meeting-free days https://posthog.com/handbook/company/culture#were-on-the-makers-schedule, and we prioritize heads down building time over perfect coordination. This will be the most productive job you've ever had.
- Ambition: We want to solve big problems. We strongly believe that aiming for the best possible upside, and sometimes missing, is better than never trying. We're optimistic about what's possible and our ability to get there.
- Being weird: Weird means redesigning an already world-class website for the 5th time. It means shipping literally every product that relates to customer data. It means building an objectively unnecessary developer toy https://posthog.com/deskhog with dubious shareholder value. Doing weird stuff is a competitive advantage. And it's fun.
WHO WE'RE LOOKING FOR
PostHog is working on a self-driving product and is training its own Deep ML models. To do this, we have a new AI Research team.
We track more products across more dimensions than any other company on Earth. Because of this, we have petabytes of data across events, errors, session replays, revenue data, and more. It’s an exceptionally large amount of data relative to our company size, and we want to use it for our users' benefit. Just imagine how, with the vast amount of data we capture, we can automatically surface insights and suggest fixes.
As an AI Research Engineer, you will play a substantial part in making this happen. This team is brand new, so you will have real autonomy, be close to customers, and greatly influence what we actually ship.
On the finance side, we have around $150M on our balance sheet, we’re "default alive," and we have a real business that has been scaling rapidly and running exceptionally efficiently with good margins for six years. We invest in what we believe moves the needle.
WHAT YOU'LL BE DOING
You'll be part of a team that researches what's next and what's better, and actually brings it into production.
Areas we’re exploring
- Session replay analysis: We already detect issues in replays, but doing so is expensive and doesn’t scale. With better models (and a huge amount of DOM mutation data), we think we can make this faster, cheaper, and best-in-class.
- User behavior prediction: Identifying drop-offs, suggesting improvements, and potentially improving conversion rates automatically without heavy manual analysis or agent overhead.
- Synthetic user testing: Simulating users to catch confusing flows or breakages before release. If this works, it could remove one of the biggest bottlenecks in modern engineering: testing (rather than writing code).
JOB REQUIREMENTS
- A PhD is a plus, but a strong background in math is a must, or you have previous experience training models (strong math skills are still required).
- Strong PyTorch experience.
- Strong ability to code in Rust, CUDA, or C (with a preference for Rust).
- Solid understanding of Transformers (not the movies).
- A product mindset. you’ll be building things!
- This is a unique role within PostHog and, as such, is a hybrid-remote position based out of our London office.
WHAT WE’RE NOT LOOKING FOR
- Pure theoretical expertise: this is not a Jupyter notebook job.
- Someone who just wants to do research and publish. Yes, you can publish, but shipping takes priority.
We are committed to ensuring a fair and accessible interview process. If you need any accommodations or adjustments, please let us know
#LI-DNI