Claudia Ng

Lesson learned: AI fails silently. Unlike traditional software, it can produce plausible outputs that are completely wrong.

Monitoring AI products means tracking behavior, not just crashes.

I built an AI to help people learn Cantonese. I forgot about it over Christmas.

API credits expired.

For two weeks, the AI returned “I don’t understand, please try again” for every query.

No errors, no alerts. Users dropped off quietly.

Thanks for featuring my post!

To all the engineers who got paged during the AWS outage last night: I feel for you.

Being on-call is rough.
Being on-call for something you can’t fix is existentially rough.

You’re out here rebooting servers at 3 a.m., praying the cloud gods show mercy.

Take a long nap today.

You don’t need to know everything.

Just enough to apply it.

If your goal is to use Python →
Build one small project that mimics your dream role.

Want to be a data analyst? Analyze a public dataset.

Want to do ML? Build a simple classifier end-to-end.

If your goal is to land a job →
Do 1–2 LeetCode problems a day for a month.

It’s boring at first.

Then one day, you realize you can solve problems you used to skip.

When I was learning Python, I kept forgetting everything.

I’d ace a tutorial… then freeze the next day staring at a blank screen.

What helped wasn’t more studying. It was more doing.

Here's what I would tell myself if I could go back in time 👇

In the past 2 weeks, 85 people tried my Cantonese AI partner:

213 conversations

993 interactions

A couple users came back 6–7 days straight

Most dropped after day 1

Now I’m stuck: Do I keep fixing bugs + adding features? Or pause and set up Stripe to see if anyone pays?

I got a “fan message” for my AI prototype!!

It wasn’t about the tech. It wasn’t about features.

It said:

“Thank YOU so much for making this. I love it. I will keep going.”

🥹

Thank you so much, Katrina! This means a lot :) What are you writing about?

“My biggest lesson was realizing that domain expertise matters more than algorithmic complexity.“ | Towards Data Science Claudia Ng reflects on real-world ML lessons, mentoring newcomers, and her journey from corporate ML to freelance AI.

@ds-claudia.bsky.social explains why domain expertise often outweighs algorithmic complexity. In this Author Spotlight, she reflects on her path from corporate ML to freelance AI, mentoring newcomers, and the lessons learned along the way.

Woke up to see AI Weekender ranked #51 in Tech on Substack 🎉

I started on Jan 23 with one goal: publish every Thursday.
Some posts hit, others sank.
But showing up each week built momentum.

Biggest lesson? Consistency compounds.

aiweekender.substack.com

“My biggest lesson was realizing that domain expertise matters more than algorithmic complexity.“ | Towards Data Science Claudia Ng reflects on real-world ML lessons, mentoring newcomers, and her journey from corporate ML to freelance AI.

✨ In our latest Author Spotlight, @ds-claudia.bsky.social, who successfully shifted from a corporate role to freelance, provides hard-won advice on navigating career paths, mentoring newcomers, and building solutions you truly own.

Every project should either:

- Build skills
- Build assets
- Build your brand

If it’s doing none, stop.

3 Solopreneur Paths for Data Scientists What I'm building after corporate DS and the paths I've tried

As a solo data scientist, I’ve found 3 types of leverage:
- Coaching gave me human leverage
- Consulting gave me applied leverage
- Content and AI tools? That’s scalable leverage

Read more here:
open.substack.com/pub/aiweeken...

I didn’t build my first RAG tool because I thought it would go viral.

I built it because people kept DM’ing me asking how to break into data science or build with AI, all questions I’d already answered in my posts.

Try it: assistant.ds-claudia.com

RAG Tutorial with Code: Ship Your Own AI Assistant This Weekend The exact blueprint for building a RAG chatbot that will save you dozens of hours

I was tired of AI tutorials that felt like science fair projects.

So I built a real RAG project: it turns your Substack posts into a searchable AI assistant.

This post walks through every step: from scraping to chunking to shipping.

shorturl.at/GicNL

Building an AI product in 5 steps:

1. Have an idea
2. Prompt like a maniac
3. Watch it almost work
4. Panic, then debug
5. Ship it anyway

Does this sound familiar, or is your chaos different?

Most of AI building isn’t about being smart.

It’s about being stubborn enough to fix what breaks…

And curious enough to try again.

Some projects take a weekend.

Some take months.

Both are valid.

Both move you forward.

From Data Scientist to AI Builder: What Actually Works How I learned to stop chasing tools and start solving problems

I don’t want to build flashy AI.

I want to build stuff people actually use.

aiweekender.substack.com/p/the-strate...

The real flex isn’t how smart your AI project is.

It’s how clearly you can explain it to a non-technical friend.

Don’t aim for perfect.

Aim for finished.

Build momentum!

I don’t build AI projects to impress anyone.

I build them to prove to myself I can.

Why do you do what you do?

Most people are waiting for the confidence to begin.
But confidence is what you earn after you start.

Your weekend build might not be world-changing.

But it might change your world:

Your confidence,

Your direction,

Your job.

I write AI technical tutorials not because I know everything,

but because writing is how I figure things out.

Genuine question: Do you consider vibe coding with AI "cheating"?

I see it like using:

- Calculators instead of long division, or
- GPS instead of memorizing maps, or
- Driving a car instead of walking.

Same destination, better tools.

What's your perspective?

Cantonese diaspora: tired of sounding like a 5-year-old when talking to your grandmother?

I'm building an AI conversation partner where you can practice without judgment.

Looking for beta testers (30-min sessions) to shape this product! DM me.