Sarah Wang

📍 Melbourne
💻 Building AI products
📚 Learning AI Engineering
🎤 Sharing what I learn
🌱 Becoming a better engineer

[Sarah Wang @sarahwang_dev]

@YvonneYaoTech 哈哈😄 谢谢 Yvonne

[Sarah Wang @sarahwang_dev]

I read a lot in my free time, all kinds of stuff, and when I come across something I really like I usually end up sharing it somewhere.

This weekend it was a children's book — one of those emotional-literacy picture books for kids, called 《我可以！》 ("I Can!").

Recorded myself reading it in Chinese on TikTok. The whole point of it is basically: whatever you're feeling is fine. Scared, angry, jealous, proud, none of it's wrong. Every page pairs a feeling with something the kid can say to themselves about it, plus a little breathing bit in there too.

What got me is the idea isn't "stop feeling that way," it's "you're allowed to feel it." If kids actually grow up believing that, they probably end up a lot better at living with their emotions later instead of fighting them.

Honestly I think I needed this book more than the three-year-old it's written for.

Curious what everyone else has been reading lately, drop your favorites below.

Also posting the readings on TikTok if you want to follow along.

我平时挺爱看书的，什么类型都看，遇到觉得好的会忍不住想分享出来。

这个周末读的是一本儿童书——那种给小朋友做情绪启蒙的绘本，叫《我可以！》，用中文录了下来发在TikTok上。

这本书的核心意思就是：不管你当下是什么感觉都没问题，害怕、生气、嫉妒、骄傲，都不是错事。几乎每种情绪都有对应的一页，配一句小朋友可以说给自己听的话，中间还穿插了一小段呼吸练习。

我觉得挺好的一点是，它讲的不是"别再有这种感觉了"，而是"你可以有这种感觉"。小孩从小就接受这个观念，长大以后大概率更能好好跟自己的情绪相处，而不是一直在压抑或者对抗。

说实话，这本书我这个大人可能比那个三岁小孩更需要。

大家最近在读什么书，也可以分享一下。

也同步发在TikTok了，感兴趣可以去看。

[Sarah Wang @sarahwang_dev]

Had a mock interview today with a senior engineer, and walked away with more than I expected.

We covered behavioral questions (his feedback on "introduce yourself": keep it short, hit the point, structure it as bullet points, polish it with strong keywords), algorithm concepts, and three system design cases — a login flow, a payment system, and a shopping page favorite/star interaction with real-time feedback. We got into caching layers, sync vs async tradeoffs, microservices vs monorepo, and how to ask good clarifying questions in an interview.

He said I did well across the board — and the one thing worth sharpening next is turning what I already know into a sharper, more compelling story when I speak about it.

One thing he said stuck with me: in a market this competitive, talent isn't the differentiator anymore — consistency is. The people who keep showing up and keep building are the ones who make it through.

We even whiteboarded a few of the system designs together — sharing a snapshot below.

今天跟一位资深工程师做了一场mock interview，收获比我预期的多。

我们聊了行为面试题（他对"介绍一下自己"的反馈是：回答要简短、抓重点、用bullet point结构化、用强关键词打磨）、算法概念，还有三个system design案例——登录页设计、支付系统设计，以及购物页面收藏功能（点亮星标、实时反馈）。中间讨论了缓存层、同步vs异步的取舍、微服务vs monorepo，还有面试里怎么问出好的澄清问题。

他说我整体表现得很不错——接下来值得打磨的一点，是把自己已经掌握的东西讲得更精炼、更有说服力。

他说的一句话让我印象很深：在这么卷的市场里，天赋已经不是决定性因素了——坚持才是。那些持续出现、持续在做的人，才是最后走到终点的人。

我们还一起画了几个system design的图，附一张截图。

[Sarah Wang @sarahwang_dev]

很多人还在问：AI 会不会取代我？
问错了。真正的问题是：
你的工作里，有多少步骤已经可以被智能体自动跑完？

聊天只是入口。
规划、调用工具、连续执行，才是分水岭。

2026 不是模型年，是智能体落地年。

[Sarah Wang @sarahwang_dev]

My career path hasn't been linear.

When I first came to Australia, one thing I wanted was simple: to become a better version of myself.

Looking back, I think I have.

I spent five years studying Pharmacy before realising it wasn't for me. I moved into education, then came to Australia to study Data Analytics — where I was first introduced to coding.

My path into software engineering wasn't straightforward either. During COVID, studying and job-hunting were tough, so I relied on self-study. After graduating, I taught myself to code from scratch and landed my first opportunity in IT.

I'm grateful to my first team for believing in me and giving me space to grow. That experience shaped who I am as an engineer today.

But there was another challenge:

Imposter syndrome.

I often underestimated myself and hesitated to put myself forward for opportunities beyond my comfort zone.

"Maybe I'm not ready yet."

Eventually, I realised waiting to feel ready was holding me back.

So I started challenging myself — not just by learning, but by building, sharing and giving back.

I built AI-powered projects around books and personal knowledge, and created AU Council Explorer to bring council and community info together for new migrants.

I took part in hackathons, including the Easygo Mini Hackathon, and recently shipped my own Chrome extension for bookmark organisation.

I also grew within the Make AI Practical community — from attendee, to volunteer, to speaker.

I've mentored students through the RMIT STEM Mentoring Program — meaningful because I remember navigating a career change without a clear roadmap.

I've also started writing about my learning on Medium.

These experiences taught me:

Confidence doesn't always come before the leap. Sometimes it comes because you took it.

My full-stack background gives me a strong foundation for building AI-powered products end-to-end, and I've realised I genuinely love turning ideas into technology that solves real problems.

Those earlier years in education and client-facing roles gave me something else I didn't expect to carry forward: strong communication skills — explaining technical ideas clearly, whether I'm pitching a project, mentoring a student, or speaking to a room of non-technical stakeholders.

My timeline may look different from someone else's. But different doesn't mean behind.

I'm now looking for my next opportunity in Full-Stack Software Engineering, with a particular interest in AI Engineering and AI-powered products.

I've been actively applying, but interviews aren't easy to come by in this market. Instead of letting that discourage me, I'm focusing on building my network, learning from experienced engineers, and continuing to build.

I'm hoping to hear from experienced engineers interested in mentorship, system design, and AI engineering.

If you're preparing for interviews and want to practise mock system design, reply below or DM me — I'd love to.

And if you're changing careers or feeling like you're not where you "should" be yet:

Please keep going. Your path doesn't have to be straight.

I'd love to hear your story — what helped you find your direction?

#CareerChange

[Javier Writes @javier_otieno1]

To create value, solve your own problems and ensure the solutions affect as many people as humanly possible.

[Sarah Wang @sarahwang_dev]

@369Serena @changloria0816 加油鸭 ~

[Sarah Wang @sarahwang_dev]

AI最近让我改观的4件事：

1️⃣ AI真的开始进入保守行业了
以前对AI进金融、能源这类强监管行业一直持怀疑态度。但最近听到的几个案例改变了我的看法——KYC自动化、能源工程工具、机场无护照通关，这些不是边际效率提升，是真的在改变事情原本的运作方式。

2️⃣ 规模化和安全并不冲突
能真正落地的AI部署，共同点都是有一套组织真正信任的、可控的AI基础设施——不是随便调个现成API，而是团队直接跟供应商合作，用自己的数据训练定制模型。

3️⃣ 小而专的模型正在悄悄成为未来
针对特定领域微调的小模型，很多时候在具体任务上表现反而超过通用大模型——更高效、更一致，而且训练它的组织自己拥有它，不被供应商锁死。

4️⃣ 不是所有噪音都值得追
一个创业者跟我说，她基本不看AI新闻，除非能直接对应到她在做的东西上——没有FOMO，也不用表演自己很跟得上潮流。这种清醒比"追热点"难得多，也重要得多。

如果你一直因为没试过某个新工具/新模型而有点愧疚——真的不用。

引用Thu Hien Vu 的分享，很赞同她的感悟。

[数字生命卡兹克 @Khazix0918]

我们经常会感觉，AI 已经这么火了，可是在我日常中，跟很多别的行业的朋友们聊，大家总是会天然的觉得，用 AI，是一件很难的事情。
有很多人，不敢开始。

包括其实我日常被问的最多的问题，也是类似的：

卡老师，我想学 AI、用 AI，我要怎么开始。
所以呢，我今天就想写一个非常非常初级的小流程，如果你想用好 AI，你就按着一步一步做，不仅可以入门，也基本有了最核心的心法。

1. 首先，你需要一台电脑，Mac 最好，Windows 也可以，内存不要低于 16 个 G，你一定需要一台电脑，这个是我们用好 AI 的前提。

2. 如果你能使用海外产品，Google 搜索 ChatGPT，然后注册账号，冲一个月 20 美金的订阅费用，再在官网上，下载 Codex，然后安装。

3. 如果你不能使用海外产品，百度搜索 WorkBuddy，下载然后登录，可以先不充会员，记得嫖一下积分。

4. 在 Codex 中把模型选择为 GPT-5.6 Sol 最高，在 WorkBuddy 中选 Kimi K3。

5. 下载一个语音输入法，Mac 可以使用豆包输入法，Windows 可以使用微信输入法，从此以后，你跟 AI 发送消息的时候，就不要再打字了，直接用嘴进行表达。

6. 把你脑海中你现在最头疼、最不想干、最抗拒的事情，一股脑的用语音输入法说给 AI，记得用【背景、痛点、需求】的框架来去说，然后再附上一句，"请你先不要开始干活，用苏格拉底提问法向我提问，一次只问一个问题，直到你认为已经完全理解这件事。"

7. 回答 AI 跟你提出的问题。

8. 等 AI 跟你问的差不多了，输出了一个最终方案的时候，把你跟这个事情相关的所有文件，比如文档、表格、PPT、参考资料啥的，放在一个新建的文件夹里，扔给 AI，然后说，“这是我的所有的文件，请你帮我直接完成吧，遇到必须由我决定的地方，再停下来问我。”

9. 调整 AI 给你的结果，千万别觉得 AI 不行然后就自己去干了，告诉 AI，哪里不对，让它去调整，如果你说不出来，也可以说 “我现在讲不出来具体原因，但我就是不喜欢，请你对比几个方向，一次问我一个问题，帮我把判断标准找出来。”

10. 大概 2～3 轮，你的这个最不想做的事，AI 大概率就替你做完了。

11. 跟 AI 说，帮我把刚才的整个过程和这个事情，做成一个可以供我反复使用的 Skill，以后，类似的事情，你就可以用 AI 自动化了。

12. 从此以后，遇到不想干的，扔给 AI，提供你的文件和【背景、痛点、需求】，让 AI 帮你完成，同时沉淀成 Skill。

好了。

现在，只需要半天时间，你已经学会使用 AI 了。

没错，就是如此简单。

[Sarah Wang @sarahwang_dev]

@yihui_indie 不错不错，后期会体验你开发的这个

[Sarah Wang @sarahwang_dev]

@randyloop 太实用了👍

[Sarah Wang @sarahwang_dev]

@tangjinzhou 谢谢回复

[Sarah Wang @sarahwang_dev]

@GeekCatX 其他agent claude 可以安装吗

[Sarah Wang @sarahwang_dev]

@GeekCatX 点star啦

[Sarah Wang @sarahwang_dev]

@GeekCatX 加啦

[Sarah Wang @sarahwang_dev]

@yihui_indie 哈哈，有道理

[Sarah Wang @sarahwang_dev]

@369Serena 过期了，如何扫码

[Sarah Wang @sarahwang_dev]

@geekplux 为啥

[Sarah Wang @sarahwang_dev]

Found a genuinely useful agent skill and sharing: last30days

/last30days <topic> sweeps Reddit, X, YouTube, HN, Polymarket, GitHub, arXiv in parallel — scores everything by real engagement, not editorial pick. Zero config to start, no API key needed.

https://t.co/a7xjtvknEU