I love to teach. I post about #dataengineering and #mentalhealth daily. 

Many years of data engineering experience doing growth analytics at Facebook, security infrastructure at Netflix, and pricing and availability infrastructure at Airbnb. 

You can find me by looking up EcZachly on all social networks 

Join my Data Learning Platform https://www.dataexpert.io

Zach Wilson

To the people who lost their jobs at Block, I want to also say, I’m really sorry. This is a very tough spot to be in! 

The DataExpert.io team is offering a free one year subscription to our cloud learning platform to anyone affected by these layoffs to help them upskill in AI and data. 

Send an email to support@dataexpert.io with your name and a brief note to get onboarded

We are going to get through these dark times together!

Founder @ DataExpert.io | ex-Netflix ex-Meta staff engineer | Angel Investor in 6 startups | Featured on Forbes | Dogs

Nirmal is a Data Scientist who yearns for solving problems that actually impact product strategy and business outcomes. As a self-described realist, he believes that with pragmatism and high-quality data, Data Science can be a very powerful tool. 

If you are preparing for ML interview, I have something to share: https://onlyoneoutlier.gumroad.com/l/decodingML 

Nirmal has experience working for both government and private industries. He had worked as Security Data Analyst for Department of Navy, and had also served in US Army. He loves working in the rare intersection between security and data science, where there are ample challenges to tackle.

If you have stalked me this far, you have to hit that follow button if you have not done yet :) Jokes a side, I love to share frequent posts on data science, machine learning, mentoring and career coaching, so following me won’t hurt.

Want to have some data science career chat regarding interviews, resume reviews or anything?  You're welcome to reserve a time at: https://topmate.io/nirmal_budhathoki

I also write data science blogs in substack, please subscribe at: https://onlyoneoutlier.substack.com/

Nirmal Budhathoki

Vibe Coding at its best.

You entered your password.

It compares and tells you - that’s wrong. 

But it is nice enough to suggest you what to enter next 😆 

#randomtopics #vibecoding #humor

Security Data Scientist @ Microsoft | Data Science, Cloud Security, Machine Learning | Mentor

graphical user interface, text, application, chat or text message

One of the biggest interview mistakes I see ...

Candidates often answering just the 'what' piece of the question. For example:

Interviewer: What is overfitting?

Most candidates say: When a model memorizes noise and doesn’t generalize well. It performs well in training, not test.

While the answer is correct, interviews are not Wikipedia tests. What they are looking for:
- How did you detect it?
- How did you fix it?
- What tradeoffs did you consider?
- What happened in the real world?

A stronger answer should immediately follow up definition without interviewer seeking for it. For example- add below on above answer.

In one of my projects, our training AUC was 0.95 but validation dropped to 0.79. Cross-validation variance was high. That’s when we suspected overfitting. We reduced model complexity, applied regularization, and added more data, that helped performance stabilize at 0.84 validation AUC.

Interviews don’t reward definitions. They reward decision-making under constraints. Find the level that separates surface knowledge from applied science.

#datascience #interviewtip

Most common resume error is not the contents but the format. 

I have reviewed hundreds of resume, and the most common mistake that I come across is not about the contents but the organization of it. And that actually matters since we are trying to simplify the ranking. 

For data scientists, your first storytelling task is your resume itself. Remember: the first thing is to pass ATS checks (complex format fails parsing) , and afterwards you have seconds to convey the message to recruiter screening. 

For me, KISS (keep it stupid simple) principle goes a long way. Below is what I recommend:

- career summary
- skills
- experience 
- education 
- projects (new grads/ entry level) 
- certifications/ credentials (if any) 

That’s all you need. 

One pager is the golden rule. 

#datascience #resume

No alternative text description for this image

I don’t usually share mentee wins to flex.

I share them for those who are still in the middle of the job hunt. And I am damn too excited not to share 😆 

This morning, a mentee messaged me:
Two offers landed her inbox 📥 

Months ago, she was unsure how to position her impact and leadership. We worked on tightening her stories, sharpening analytics depth, and running mock interviews tougher than the real ones. She gave 200% so she deserves this win. 

I do take mentorship seriously. When someone books a call, I prepare, I challenge, and I push. You paid for it, trusted me, so your career deserves that level of respect.

If you’re in the process right now:
Rejections are not the verdict. Iteration is.

Keep refining. Your breakthrough might come in multiples.

I dont preach for zero to hero deals, I let you become your own hero. 

#datascience #mentorship

The guy who retired from Microsoft to chose goose farming knew what's coming 😅 

If you remember one viral post a while back about a microsoft employee, who after 22 years, retired to chose goose farming. While I brought this for a funny context, but we cannot deny that AI is coming after many digital jobs. 

Anthropic just released a fascinating labor market report. While the radar chart is beautiful, sometimes bar chart do a simple job of better explanation. And who else to ask to convert the radar into bar- Claude itself. 

They introduced a new metric called observed exposure: not just what AI could theoretically do, but what it's actually doing in real workplaces.

• Computer & Math: 94% theoretically exposed, 33% actually observed 
• Physical roles (farming, construction, food prep): lowest on both

The red bars (observed usage) are a consistent fraction of the green bars (theoretical capability), hinting that the ratio of "actual to possible" is somewhat uniform: the gap isn't just a few laggard sectors, it's everywhere.

🔍 On jobs, the early data is more measured than the headlines suggest:
• No significant rise in unemployment for highly exposed workers (yet)

AI's theoretical reach is vast, but real-world adoption is still a fraction of what's possible. However this gap between capability and usage is narrowing. Most of these roles will be evolving in skillset. Therefore- staying back being skeptical about AI - is NOT an option.

Worth reading if you're thinking about how AI will reshape work over the next decade. Full report link shared on comments.

#datascience #AI #jobmarket

The Data Science Trap No One Talks About

Everyone wants to learn:

	•	Python
	•	Machine Learning
	•	Deep Learning
	•	LLMs
	•	GenAI
	•	Agents

But almost no one talks about:

	•	Problem framing
	•	Stakeholder communication
	•	Metrics definition
	•	Tradeoff thinking
	•	Business impact

That’s the difference. 

And guess what AI is good at ? 

Let that sink in.

#datascience #ai

I’m excited to announce that I have successfully completed …

Vacation with Kids: Prod Deployment

No dev or ppe 😂 

Scope included:

- Airport sprints with emotional baggage

- Snack demand every 17 minutes

- Are we there yet? , lets talk about the real-time monitoring system

- Many high- priority incidents reported by kids, thar we quickly de-prioritized 

There were delays. There were escalations.

But we achieved the primary KPI:
Maximise the memories created.

We talk a lot about grind, growth, chasing the next milestone.

But the ROI on family time? It compounds quietly.

The chaos fades. The memories don’t.

Taking a break is not a weakness. Its a fuel for your productivity. 

Back to grind now 😆 

#randomtopics #family #break

From Student to Full Time Employee 

Everyone’s journey is unique, everyone’s struggle is different, but a little help from someone can go a long way in your career. 

If you are in asking position today- be sure to be on the giving side tomorrow when you can. 

Last year, we had few career conversations.
He was still a college student, figuring things out, learning about job market and preparing the best for it. 

Today, he messaged me saying he received an offer from Microsoft for a contract Data Science role in Vancouver.

I appreciate how he remembered to send a gratitude, he could not skipped, but he did not. While I dont expect my memtees to thank me , but it does feel special to be tiny part of their success stories. 

You never know how few mins of call can impact someone’s trajectory.

So far this year has been amazing. I am hearing back from folks about offers, interviews, breakthroughs.

Mentorship is not about a single call. It is about planting seeds. And sometimes you get to watch them grow 🌱

To everyone still in the process, especially students wondering when it is their time to celebrate:

Keep going. You never know how close you are until get there. 

#randomtopics #mentorship

undefined undefined

You will always be my guru …

Guru here means mentor. Some messages hit differently. 

Today I received this note from someone I mentored a while back. She took the advice to pursue a Master’s in Data Science, and just landed a Senior Data Scientist role in the biotech industry.

Moments like this remind me why I keep mentoring.

Not for recognition.
Not for visibility.

But because helping someone change the trajectory of their career is one of the most meaningful things you can do.

To everyone currently learning, struggling, applying, and doubting yourself- keep going.

And to my mentee- I am incredibly proud of you, and thank you for your kind words. 

Your hard work paid off 🚀

#datascience #mentoring

From  “unfortunately” to  “congratulations” …

My inbox has so many emails that start with one word.

“Unfortunately…”

Unfortunately, we will not be moving forward.
Unfortunately, we chose another candidate.
Unfortunately, we found a closer match.

Well it sucks to get those, but you know what doesn’t-  getting the Congrats email. 

Every ‘unfortunately’ was actually training.
Training in patience. Training in resilience.

Success is rarely built on a straight line of yes. It is built on a pile of no’s, quiet persistence, and the belief that the next door might open.

Those emails did not define me then. They do not define me now. And they definitely will not define tomorrow.

If your inbox looks like this right now, keep going. One day it will change …

🔜

#datascience #jobmarket

I’ve worked in the DoD.
I now work in Security and AI.

So when conversations around AI safety, defense use, and governance come up: 
I don’t see it as black and white.

AI will absolutely support defense missions: cyber defense, logistics, threat detection, decision support. That’s not speculation. That’s reality.

But with powerful models comes responsibility.

The company that builds and owns the model must define and own clear governance, usage boundaries, and safety guardrails. Policy cannot be an afterthought. 

On this, I respect Anthropic’s stance and leadership from Dario Amodei. 

You may agree or disagree with specific decisions, but taking model governance seriously deserves acknowledgment.

This isn’t about being pro- or anti-defense.
It’s about being pro-accountability.

AI will shape national security. No doubt. But the real question is whether we build it responsibly.

As we are now stepping into another war, the use of AI into the weapon systems without national level policy can be extremely dangerous.

#ai #safety

Microsoft confirmed that Anthropic’s Claude will remain available to most customer, even as DoD labeled Anthropic as ‘supply-chain risk’. The restriction applies only to the Defense Department.

At the core of the tension is Anthropic’s stanc on limiting certain uses of its models, such as mass surveillance and autonomous weapons.

This highlights an emerging reality in AI:
The debate is no longer just about model capability, but about who defines the boundaries of how AI should be used.

#datascience #responsibleAI

Sharing the recording from our recent AMA session on DS Interviews

We recently hosted an AMA session on Data Science and AI careers, and the conversation was packed with practical advice from industry experts.

If you are navigating the data job market, a few key reminders from the discussion:

- Your resume is the door opener. Tailor it for the role you are applying to.
- Impact matters more than tools. Showcase what problems you solved.
- Build projects that demonstrate real thinking, not just tutorials.
- Start networking early and genuinely. 
- Be prepared for AI and ML interview questions beyond just definitions.

We also discussed career transitions from analyst to data scientist, portfolio building, and how the job market is evolving.

Thank you again to all the participants, and special thanks to my panelists: Karun Thankachan , Shivam Bindal, Sai Kumar Bysani.

As announced during the session, our three random lucky winners for the Decoding ML Interviews book are:
Sujata Biradar
Lance Hester, PhD EE
Maxime Chung

For those who could not join live, sharing the full recording below.
[https://lnkd.in/gRDRuYXE]

Hope it helps someone preparing for their next step in data science.

#DataScience #AI #CareerAdvice #InterviewPrep #MachineLearning

Data Science & ML Interview Insights

Long post but please bear with me, it has useful info. Recently we completed our AMA session on data science interviews. This is based on what we covered. Sharing for folks who missed. 

🔹Strategic Career Transitions: Transitioning from specialized domains (e.g., Energy, supply chain): Do not start from scratch; bridge the gap by applying ML to your domain-specific problems. Focus on implementation part, deploying a simplified version to show you can bridge the gap between theory and practice.

🔹Navigating Gap Years: Volunteering projects are highly valuable if they show continuous skill application. In interviews, proactively lead with the impact of these volunteer roles to normalize the gap. These days some startups and nonprofit organizations offer volunteer roles. 

🔹MAANG Interviews: Most tech companies often emphasizes general cognitive ability and statistical first principles. While LeetCode is less common for product DS roles, DSA is still tested for Applied DS and MLE rounds. Most companies do easy to medium level, but some roles in MLE can go beyond that. 

🔹The Evolution of ML System Design: Modern ML System Design now includes "Context Engineering" and "AI Evals". Interviews are shifting toward how you evaluate LLM outputs and manage RAG retrieval quality rather than model architecture. Focus on quantization, model compression, and latency optimization. Companies look for projects that show you can run models under strict hardware constraints.


🔹Redefining "Impact" for Students: If you lack a real business, create a "Synthetic Stakeholder". Define a clear ROI goal for your project (e.g., "This fraud detection model aims to reduce false positives by 10%, potentially saving a mid-sized bank $X").

🔹The Importance of Thesis Work: For Masters students, a thesis is critical if you are targeting Research Scientist roles or PhDs. For general industry roles, a strong, deployed portfolio project often carries more weight than a purely theoretical thesis. Skills matter the most. Relevant courses, and your projects weigh far more than the exact name of your degree (e.g., Industrial Engineering vs. Data Science).

🔹Handling Large Datasets in Portfolios: You do not always need millions of records. A well-cleaned 30k–50k dataset that allows for complex causal inference or A/B testing analysis is often more impressive than a "messy" massive dataset with shallow analysis.

🔹Some of the recommended resources: 

- A/B Testing: "Trustworthy Online Controlled Experiments" (Ron Kohavi ) 

- System Design: "Machine Learning System Design" (Chip Huyen).

- Causal Inference what if (Miguel)- Harvard has free online version- link on comments. 

- SQL/Product Prep: Stratascratch, DataLemur, InterviewQuery, and Exponent. We can also host our own DB and learn some data engineering- check out aiveen. 

- ML interview prep: we wrote a book on interview focused guide: link on comments. 

#datascience #interview

🎙️ Just released my first podcast of the year ...

It has been a while since my last episode, but I’m excited to bring the podcast back with more insightful conversations this year.

In this episode, I sat down with Riya Gayasen, a seasoned product manager, to talk about the intersection of AI and Product Management. Ironically, my best part of the session was:

👉 Great AI products start with human stories.

Riya shared her journey from engineering to product management and introduced a simple but powerful framework for integrating AI into products:

• Data Strategy: understanding your data and mitigating bias
• Feature Selection: choosing AI features that solve real human problems
• Human-Centered Design: keeping empathy at the core of product decisions

We also discussed:
- Why empathy matters in AI development
- How AI can help shape product roadmaps
- Avoiding the trap of building AI features just because they are possible

One quote from the episode that really stuck with me:
'Think human first, then scale with AI.'

If you are working at the intersection of AI, product, and data, I think you’ll enjoy this conversation.

🎧 Link to the full episode: [https://lnkd.in/gd_5b5eA]

And this time I promise- more episodes are coming your way.

#datascience #ai #podcast

Is there a curve that quietly shapes your entire data science learning journey ?

Yes, its called exponential decay (increasing form)
y = C ( 1 - e-kt ), k > 0

C: max value to achieve- where you want to be
k: your learning rate
t: time (experience)

At the beginning of your data science journey:
-Effort feels huge.
- Learning feels fast.
- Progress feels visible.

You learn: Pandas, SQL, Regression, ML algos, Sklearn, and suddenly you feel powerful. Then something strange happens. Your effort keeps increasing. But visible progress slows down.

Why?
Because early gains are steep. Later gains are subtle.

The first 6 months: You go from 0 → 60%
The next 3 years: You go from 60% → 80%
The next 5 years: You go from 80% → 90%

And that last 10%? That’s where real learning lives. We will never hit the max, as there are always things to learn.

Here’s the psychological trap: Beginners quit when the curve flattens. Seniors survive at the plateau. The difference is not intelligence. It’s patience.

Weekend reminder:
If your progress feels slower than before, you could still be learning. The curve flattening is not failure.

#datascience #learning

The hardest part of data science that no one teaches you ...

After my last post about technical vs. business skills in DS: someone asked me-

Okay… but how does a new grad build business skills without actually working in industry?

Fair question. I get that a lot, so I decided to share here. Here is my take on it.

You can’t simply wait x years to learn this. So here are 5 ways on- how to build “business thinking” before you have a big title.

1️⃣ Turn every project into a decision: Instead of saying: I built a churn model. How about saying: This model helps the retention team decide who to offer discounts to. Redefine your problem statement to cover the business aspect. Even though you take the dataset from Kaggle , you are welcome to reframe the problem into business language covering:

- Who is the decision maker?
- What decision are they making?
- What happens if they are wrong?

2️⃣ Read product docs also, not just ML blogs: Instead of only reading about transformers and diffusion models, read:
- Product case studies
- Earnings reports
- Experiment write-ups
- Growth breakdowns
Once you learn how companies talk about metrics, you will find out how everything revolves around tradeoffs.

3️⃣ Reverse engineer the metrics: Pick any app you use daily. Ask:
- What is their north-star metric?
- What behaviors drive it?
- What metric would I track weekly?
- Where would ML actually help?

The same app you loved so much, you'll see some critical feedback. The same app you hated before, you'll see some positives. Your thinking will change.

4️⃣ Practice explaining to a non-technical friend: This is a good one. Can you explain the most technical concept to the least technical person. Interviews also test this skill. Example: If you can’t explain your project without saying: "feature engineering” or “hyper-parameters”, you haven't understood the impact yet.

5️⃣ Always think in terms of tradeoff: For each project, ask yourself:
- If I improve precision, what suffers?
- If I reduce fraud alerts, what risk increases?
- If I automate this, what human judgment do we lose?

The biggest myth: You’ll learn business once you get a job. 

Lets bust this myth, shall we ?

#datascience #businessimpact

AI Meets Cybersecurity …

And markets react. 

One of the reasons I love Anthropic : security has never been their secondary option. 

They just unveiled Claude Code Security, an AI-driven feature that scans entire codebases for vulnerabilities and suggests fixes with human oversight. It’s part of a broader push to bring generative AI deeper into security workflows.  

The market’s response was swift: major cybersecurity stocks like CrowdStrike, Cloudflare, Okta, Zscaler and others saw sharp declines this week as investors priced in the possibility that AI automation could compress demand for traditional security products.  

While this could just be a immediate reaction, but one thing is sure-  security for AI will be next big thing, and if these companies only think for traditional security features then they will left behind.

Interesting read- blog link on comments. 

#datascience #ai #security

This post is not about data science today.

It’s about something more important: how well we are prioritizing our work vs. our wellness. My wife works as a nurse, and she recently mentioned something that caught my attention: Her urgent care is more patients with stress-related health issues as compared in past. And there was a pattern she noticed: many of them work in tech.

- Maybe it’s the job market uncertainty.
- May be its a series of layoff and rejections.
- Maybe it’s the long hours and deadlines

Recently I also came across some very sad news within my own community, which made me pause and reflect.

Your career matters. Your ambitions matter. Your goal matters. But your health matters above all. Someone once told: you can have 99 problems until you have a health problem, then you have only 1 problem. 

No job, promotion, or deadline is worth silently carrying stress that breaks you over time.

- Take the break
- Talk to someone 
- Log off when you need to

Success should not come at the cost of your well-being. If you’re feeling overwhelmed lately- you’re not alone. And it’s okay to slow down.

I am happy to schedule sometime for free if you feel that I can be some help.

#yourhealthmatters

Goodhart’s Law: The Metric Trap

When you are obesssed with metrics, you tend to work too hard for it. But does it solve the real problem ? 

In data science, we often optimize for a proxy metric, but systems and people quickly learn to game them.

The Infamous Example: The Cobra Effect

During British rule in India, the government was concerned about the venomous cobras. They offered a bounty for every dead cobra brought to them.

• The Goal: Fewer snakes.
• The Metric: Bounty per dead cobra.

Result showed tremendous on number of cobras killed. But what they didn’t t realize is that people were breeding cobras for cash. 

When the bounty ended, they ended up leaving more snakes than they started with.

Some examples of modern data traps:

- High CTR: Often leads to clickbait, not quality.

- Lines of Code: Leads to bloated, inefficient software.

- Accuracy: On imbalanced data, 99% accuracy can still mean a useless model.

The Fix: Counter-Metrics

Never track a target in isolation. If you optimize for conversion, monitor Refunds. If you optimize for speed, monitor error rates. If you are interviewing for product DS roles, make sure to add guardrail metrics. 

Models are literal. They’ll give you exactly what you ask for, even if it’s not what you actually wanted.

P.S. Even this blowhole selfie is a lesson in Goodhart’s Law- I optimized for the perfect ‘blow’ and nearly lost my phone.

#datascience #ai #metrics

taplio

The Battle

Zach Wilson

Nirmal Budhathoki

Followers

Posts

Engagements

Likes

Performance ScoreZach Wilson's LinkedIn Performance

Performance ScoreNirmal Budhathoki's LinkedIn Performance

Ready to grow your LinkedIn presence?

taplio

The Battle

Zach Wilson

Nirmal Budhathoki

Followers

Posts

Engagements

Likes

Performance Score.css-goub6s{font-size:var(--chakra-fontSizes-md);font-weight:500;color:var(--chakra-colors-gray-600);display:block;margin-top:var(--chakra-space-1);}Zach Wilson's LinkedIn Performance

Performance ScoreNirmal Budhathoki's LinkedIn Performance

Ready to grow your LinkedIn presence?

Performance ScoreZach Wilson's LinkedIn Performance