Microsoft

Currently working as Data Scientist in Microsoft Security. 
- Building anomaly detection model using unsupervised techniques on user or application service behavior and activities based on identities used for login and access of resources.
- Formulating problem statement, gathering project requirements, and defining success criteria, in collaboration with product managers and other stakeholders. Data-> Modeling journey is incomplete without business understanding. 
- Building data pipeline and ETL jobs to bring multiple data sources into integration, and create aggregated views as a part of feature engineering.

Senior Data & Applied Scientist

Great Learning

Conducting the mentored learning sessions for the cohorts of Data Science and Machine Learning certification program from MiT Institute for Data, Systems, and Society (IDSS) in collaboration with great learning. 

Link to the program: https://idss.mit.edu/about-us/

I also started to mentor the post graduate certification program on AI and ML for business applications with UT Austin.

Link: https://www.mygreatlearning.com/pg-program-online-artificial-intelligence-machine-learning

Data Science Mentor

VMware Carbon Black

Data Scientist at Security Business Unit of VMware Carbon Black. 

- Built prevalence-based scoring model that effectively suppresses the noise (false positives alerts) by x% , saving the time and resources for threat analysts and hunters
- Worked on building an unsupervised machine learning model to cluster the command lines, identifying anomalous command lines that can result in malicious behavior (algorithms explored in this project are- Doc2Vec, UMAP,  K-Means, and HDBSCAN)
- Created data transformation pipeline in Apache Airflow to orchestrate and schedule the jobs for batch preprocessing, which leads to effective feature engineering for training ML models
- Designed architecture and threat models, in collaboration with engineering team to productize ML model
- Worked with ML engineers to understand, establish and follow CI/CD pipeline for MLOps

Senior Data Scientist

Worked as a Data Scientist for Azure Core Security Service team under Microsoft Cloud + AI.

- Worked on detection models to find patterns, detect advanced threats, bad actor techniques, anomalous or suspicious activity, and active risks to Azure network and systems
- Conducted end-to-end data analytics and reporting on UDP based DDoS attacks using unsupervised learning methods like clustering and pattern analysis to detect and fix the vulnerabilities like OS baseline configuration, open ports/protocols in network security groups
- Performed exploratory data analysis on internet exposed endpoints, curate the raw data from different sources for machine attribution, and generate reports providing actionable insights
- Created a baseline model of Incidents data to help calculate and optimize mean TTA/ TTM (Time To Acknowledge, and Time To Mitigate) metrics for security incidents, meeting the SLAs on those metrics, and closing gaps based on root causes like ticket routing/transfer issues
- Expanded the knowledge of data science within the team by mentoring data analysts to help them learn/grow in the data science domain, and collaborated with data engineers/developers to create, manage and monitor the data pipeline for the ETL workflow

Data & Applied Scientist

Wells Fargo

Worked on building machine learning model for Credit Card debt collections to predict whether customers will contact and pay once their account is delinquent. This model scores the conditional probability based on historical contacts data of each customer across different communication channels like phone call, email, mobile or desktop app, branch visits, and text data. Completed data ingestion phase- working along with data SMEs from various departments to collect and ingest the training data, sampling the data and defining the base population. Worked with data engineers in the team to build data flow diagram, entity relation diagram; and use those artifacts to streamline the data pipeline and complete data transformation/validation.

Tools & Technologies: Python, Git, SQL, Jupyter notebook, Pyspark, H20 ML, SkLearn, SAS

Data Scientist

NAVWAR

-Collecting software scan data from network and process it to identify and remove vulnerable & non-compliant applications using data analytics and modeling.
-Network traffic data analysis (approx. 10K endpoints streaming gigabytes of data everyday in Research Development & Test Environment): cleaning/sampling the data, creating predictive models to categorize, block and remediate the targeted attacks and threat vectors.
-Work with security operation teams like forensics and incident management to collect required data, and perform analysis to aid their investigations.
-Understand/ follow policies regarding information/cyber security like NIST, FIPS, RMF, FedRamp and others.

Tools & Technologies: Python, PostgreSQL, JIRA collaboration tool, Splunk log analytics, Vulnerability Management System (VMS), Enterprise Mission Assurance Support Service (eMASS), Host Based Security System (HBSS), Assured Compliance Assessment System (ACAS)

Security Data Science

US Army

US Army Veteran

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.

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 weekly newsletter in substack, please subscribe at: https://onlyoneoutlier.substack.com/

Someone said it right: Google always plays for a long game ...

If you thought Google was quiet or slow in AI, they are already proving that wrong. They weren’t slow; they were steady. And steadiness always wins long races. 

In my opinion, Google is the long-distance runner in AI and will lead for a few key reasons:

🔹 No compute dependence: They don’t rely on Nvidia. Their TPU ecosystem gives them control over scale, efficiency, and cost.

🔹 Huge enterprise and consumer footprint : Google Workspace, Search, Maps, YouTube, Chrome, Android—they already have one of the largest user and enterprise customer bases in the world.

🔹 Many product surfaces for AI: Every product can serve as a distribution channel. AI upgrades can launch across billions of users overnight.

They’re building models for themselves. Not just for the market but for their own infrastructure, ads ecosystem, cloud platform, and internal workflows. That’s a barrier few can match.

Everyone notices the noise today. But long-term growth comes from scale, infrastructure, and execution. #Google has all three.

Microsoft and Amazon will get edge from democratizing AI in the cloud (AI as a service) , and that's what they are doing.

#randomtopics #airace

Nirmal Budhathoki

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

Happy Friday! A small achievement that truly made my day.

I started getting pings from a few folks at Microsoft asking about a certain program, and that is when I found out: MIT’s Applied Data Science Certification Program featured me in their official brochure. 

During my free weekend mornings, I take the mentorship classes, this program is run by Great Learning. It is one of the widely recognized certifications in Data Science, and every cohort reminds me why I enjoy helping others grow in this field.

Thank you to MIT and Great Learning for the recognition, to the learners who trust me, and to my network for always encouraging me to keep improving.

Here is to more learning, more teaching, and helping the next generation of data scientists advance in their journey.

#datascience #MiT #greatlearning #appliedAI

Statistics vs. Coding across the DS job titles. 

Where it comes to data science job titles, it varies widely, and so do the skills required for the role. How much math/stats are needed vs. how much coding is needed are the two skills always in discussion. 

I thought a scatterplot will help us visualize it. Scale does not matter here, coz its relative. The idea is to understand these skills demand based on role.

1. Research Scientist: They are ones doing more research-oriented projects applying ML. Most of the projects are not aimed for production solutions.

2. Data Scientist: They are ones doing general data science projects - it could be building ML model, building statistical model, or sometimes simply helping product team with data mining and insights.

3. ML Engineer: They are the ones to help with ML infrastructure to deploy models into production. Coding is high demand skill. Statistics is a plus. MLOps is their superpower. 

4. Data Analyst: They are ones doing data collection, analysis, statistical modeling, A/B testing, and visualization. 

5. BI Analyst: They are the ones gathering business and technical requirements of a projects; preparing reports and documentation. These roles are slowly getting obsolete and replaced by product managers as PMs or re-branded as BI Engineers. 

6. Data engineer: They are like software engineers but focused on plumbing data pipeline - basically preparing and getting the data ready for data science. Simply put: If data is not flowing, model is not building. 

7. Applied Scientist: In a way they are the most decorated when it comes to skillsets. They are ones to build Proof of concept to the model development to the model deployment. They need to have math/ ML/ plus DevOps. 

Just one favor: please share this with you network if you find it useful♻️ 

Also, this is just one of the 30 DS tips that I have shared for FREE download in my topmate [https://lnkd.in/g_MzN7eM]

#datascience #machinelearning #tips

li_at

JSESSIONID

John Sule

“ Hi Nirmal, I got the job.” 

This is one of the best messages to receive as a mentor.

The hard work, discipline, and consistency belong entirely to her. My role was just a small part of the journey- a guiding hand, not the driver.

And that’s exactly why I share moments like this.

Not to praise myself.
Not to take credit.

But to encourage those who are still in the market, still interviewing, still waiting for that one “yes.”

It will come.

Sometimes the timing takes longer, sometimes the path feels harder, but every hour you put into preparation is compounding behind the scenes.

If you’re reading this and feeling discouraged- keep going. Your “I got the job” message is closer than you think.

At the end, its not about the followers, but the hearts you touch ❤️ 

#randomtopics #mondaymotivation

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

If you are not in premium …

you are missing out 😆 

If chatGPT can fix em dashes, you can do it. 

#randomtopics #humor

Can you use PCA in anomaly detection ?

Simple question, but it clearly tests you if know the PCA more than just for dimension reduction.

Why it works ?

PCA captures the main structure of your data by learning directions with the highest variance. Essentially, it learns what “normal” looks like. If a point doesn’t align with that structure, PCA cannot reconstruct it well, and that gap becomes the signal.

Two common ways to identify anomalies:

🔹 Reconstruction Error: Low-dimensional projection → reconstruction → measure the difference. High error = potential anomaly.

🔹 Distance in PCA Space: After PCA, normal points cluster. Points far from this cluster may be potential anomalies. 

Quick example: In network traffic data, most sessions follow predictable patterns. A session with extremely high outbound bytes and low inbound bytes won’t align with the learned PCA structure. Its reconstruction error will spike, and you can flag it for further investigation/ review.

Follow- up: When do you think it will struggle?

While PCA can be used, it depends on the problem/ data. It can struggle:
- Non-linear patterns
- High-dimensional sparse data
- Anomalies aligned with major variance directions

In interviews, focus on explaining the intuition, not just the math. That’s usually what they’re looking for. 

We have covered topics like this and more in our Decoding ML book, which is currently on 20 percent discount- use the code: THANKSGIVING20

[https://lnkd.in/gWGNMyt5] 

#datascience #machinelearning #PCA #interview #decodingML

Taking a minute to - Thank all of you, who follow me, support me and/or Inspire me. 

Today I’m feeling grateful, not for the number next to my name, but for the people behind it.

In 2025, I had the privilege of being part of the journey for 11 mentees who landed their dream job. They did the hard work. But I was a small support system in their corner.

And today, I also saw something unexpected: Favikon ranked me among best content creators in the US.

The ranking itself does not matter to me. What matters is being on the same list as creators I genuinely learn from and respect.

To everyone who follows, shares, comments, sends a kind message, or recommends my content to someone who needs it: thank you. You’re a big reason I keep going.

Mission is not complete. If I can help change even one career trajectory with a small nudge, that’s motivation enough.

#datascience #content #linkedin #thanksgiving #thankyou

All that matters is the Try …

In tech and in life, success or failure is just a byproduct. What really matters is the try.

Think about how we write code.

Every error-handling block starts with one thing: try

The first step is never catch- it’s always try.

You run the code.
If it fails, you debug.
You fix the logic.
You try again.

That’s how learning works. That’s how growth works. That’s how careers are built.

Failures don’t define you. Retries do.

Keep trying. Keep iterating. Keep improving.
Everything you want is on the other side of “run it again.”

#randomtopics #job #failure #rebuild

Not all the gifts we receive come in box 📦 

Something special for this thanksgiving came in pencil and paper from my daughter- reminding me that being present, being patient, and being kind is what kids remember most.

Grateful for family, for these little moments, and for the real meaning of “thank you”
🧡

#randomtopics #thanksgiving

If only learning at school or Kaggle was enough 😆 

Real world problems come in different shapes and forms, often with lots of challenges and ambiguity. 'This data make sense' will go out of the window.

My suggestion is to find diverse set of datasets beside Kaggle while you learn, and do projects for your portfolio.  

My Top 10 besides the Kaggle: 

📌 Microsoft Research Data Set: https://lnkd.in/eic-WUbQ

📌 Azure Open Data: https://lnkd.in/g6J-zXxb 

📌 Amazon Data Set: https://lnkd.in/eRdduaaV

📌 Google DataSet search: https://lnkd.in/egyh_MtM

📌 Pew Research Center posts lots of survey data and other research data:
https://lnkd.in/gUVS-8kS

📌 Paperswithcode- the research papers are shared with code repo, often with datasets too: 
https://lnkd.in/ghpgxEbH

📌 HuggingFace Datasets: https://lnkd.in/gurQdXFB 

📌  OpenML: https://lnkd.in/gPMMA6fi type=data&sort=runs&status=active

📌  Data Gov: https://data.gov/ 

📌  Awesome dataset repo: https://lnkd.in/gtJuCrag

If you find this post useful, please share with your network ♻️ 

Follow Nirmal for more contents in data science and machine learning. 

#datascience #datasets #machinelearning

"Leaders eat last."

One of the best parts of military culture- and something that has stayed with me long after I hung up the uniform.

During Thanksgiving, leaders serve the meal in the dining hall. I still remember those moments… tired, cold, busy, but surrounded by people who became family. When you are serving, your buddies are not just coworkers- they’re your support system, your community, your home away from home.

As we celebrate Thanksgiving, I want to wish everyone a warm holiday. 

Especially those who are on duty today- serving, protecting, and standing watch while the rest of us enjoy time with our families.

Thank you for what you do.

Happy Thanksgiving 🙏 🦃 🇺🇸 

#randomtopics #happythanksgiving

The hardest season of your life often becomes the strongest part of your story.

When I first came to the US for my master’s, everything felt new- 

new culture, new pressure, new dreams. 

I worked in a small deli on the side hustle. 

While every day felt busy, there was a time that truly tested me:

Tuesday night: close the shop at midnight.
Wednesday morning: open it at 6 am. 

I worked till 5 pm, no car so walked 15 mins in freezing winter to make evening class. 

I used to complete my assignment after my shift the night before and email it to a friend to print because the professor wanted hard copies.

And ya the reality check- we did not have our ChatGPT friend 😆 

Looking back now, that period didn’t break me. It actually built me. It taught me resilience I still carry today.

If you’re going through your own version of that walk- in the cold, or heat, tired, unsure, and felt lost- please know this:

Suffering makes the luxury definable. Struggle shapes the strength you’ll need later. Your future self is being built right now.

Keep going. Your story is just getting started.

The only reason I share it- there are many of you going thru similar moments. My message: everything gonna be alright. 

The suffering of climbing makes sense once you witness the beauty from summit ⛰️

#randomtopics #mystory

No alternative text description for this image

🔥 5 Most Common Coding Patterns for Data Science Interviews (Free PDF)

DS coding interviews are always confusing. How much to prepare… and what to prepare?

Most people jump into random LeetCode questions, but the truth is this: Almost every DS coding question falls under a handful of patterns. If you master these patterns, you’ll be more comfortable tackling them. 

So I put together a simple, easy-to-follow PDF guide:
 “Five Coding Patterns For DS Interview”- Hashing, Sliding Window, Two-Pointers, Slow/Fast Pointers, and Merge Intervals. 
 
I've also created a playlist of 30 DS coding questions that will optimize your prep without drowning you in the complexity. I love #NeetCode
[https://lnkd.in/eeu3Y9r9]

While data science coding isn’t the same as SWE interviews, if you're targeting ML-focused roles, you will still be tested on coding fundamentals.

If this helps you, feel free to share it with someone preparing for their next interview. Let’s make DS interview prep less confusing and more strategic.

Follow Nirmal for more contents like this. I am always helping people to break into into Data Science & ML with clarity and structure.

#datascience #machinelearning #coding #interview

Five Coding Patterns for DS Interview

Thank you for being a part of our 'Decoding ML Interviews' Journey:

Over the past few months, many of you have been following my content, and the entire journey behind writing my first book: "Decoding ML Interviews". I just want to pause for a moment and say thank you.

Writing this book was never about adding more theory to the world. It was about giving candidates a clear, structured, and organized way to learn, practice, and prepare for real ML interviews. Thank you Karun for co-authoring it. 

The response so far has been amazing. From messages saying the book helped simplify difficult concepts, to people landing offers after using the frameworks inside- every note means a lot.

If you haven’t grabbed your copy yet, we’re currently running a limited discount for the next 50 copies- 20% off. Use the code: THANKSGIVING20

[https://lnkd.in/gWGNMyt5] 
 
This is our small way of saying thank you to the community that has supported us from day one. Let’s keep learning and growing together.

If you think your network will benefit of it, please share.

#datascience #interviews #decodingML

AI itself is not the bubble.

The bubble is in the spendings, the unchecked investments, round tripping, unclear ROI, and the rush to pour billions without a business case. Fear of missing out is real 😆 

This pattern isn’t new though. Every major revolution, from industrial, internet, mobile, it went through the same cycle. The underlying technology was real. The frenzy around it created the bubble.

And even if a correction comes, the top companies won't disappear. Their diversification across cloud, devices, ads, chips, and services makes them structurally resilient, not hype dependent.

Peter Thiel cashing out some Nvidia shares is just another spark in this over-heated narrative, but it doesn’t change the long-term trajectory of AI. 

The value is real. The business discipline around it still needs to catch up.

#AIbubble #randomtopics

Not all rejections are about you.

I spoke with a few mentees recently who did everything right.

Cleared round after round.
Strong feedback.
Full loop done.

Then came the email: Business priorities changed. No headcount. Sorry 😞 

Failing interviews is part of the process.
But passing everything and still losing the offer hurts more.

Now add visa timelines and immigration pressure, a stress many don’t have to carry.

If you don’t have to fight that battle, know this: you already have an advantage.

This market is tough. For some, it’s tougher.

If you are in a position to offer any help, please do. Its all about the community. 

If you’re still showing up after this, your resilience deserves more credit than any offer letter.

#randomtopics #jobmarket

That struggle, a true data scientist can only understand 😁 

Which conversation is harder for you?
A. Explaining that you don't fix computers as a Data Scientist
B. Explaining that you cannot just "make the blue bar bigger' by tomorrow

Honestly, I will take the cookies any day 🍪 

#datascience #humor #communication #stakeholder

What’s the worst black box model out there?

Worse than neural nets, transformers, or any complex LLM?

The interview process.

No transparency. No gradient updates. No feedback loops.

You just get … unfortunately we’ve decided not to move forward. At this time, we are not able to give any feedback. 

Now you are lost in a puzzle: what exactly should I fine-tune for next one ? 

One thing I learned the hard way: mock interviews are the most underrated part of interview prep.

You don’t know what you don’t know until someone puts you in the hot seat, asks the right questions, and shows you the gaps you never realized you had.

I committed this mistake in my early career- prepping alone, assuming I was “good enough,” and then getting blindsided on interview day.

If you’re serious about leveling up for data science, ML, analytics, or product interviews, I’m doing 1:1 mock sessions where I give detailed feedback, structure corrections, and targeted improvement plans. And I take these calls seriously, I spend time to do research to give you the best and closest experience. 

My profile has topmate link for anyone who wants to learn more. 

Let’s stop treating interviews like black boxes and start treating them like models we can actually improve with feedback loops. 

#datascience #interviews

graphical user interface, text, application

If you're preparing for Product DS role, make SQL your best friend. 

If the role is ML focused- it will be different game plan. The focus will be more on python coding, ML, Stats and ML system design. Detailed post -coming soon. This post is about product DS role, mainly SQL portion. 

Lately, I’ve been helping candidates prepare for Product Data Scientist interviews (Meta-style), and SQL is always one of the biggest differentiators. 

Not only logic and correctness are important, but how fast you get to finish line. Speed matters because there will be 2-3 questions, and SQL is only the half of the show, remaining half is product sense. If you spend too much time on SQL, you are killing your own clock for case study.

Here are two realistic SQL questions I used in a recent mock interview (Meta-style). I’ve attached the schema image. Reels are short videos in Instagram. It can be surfaced via two ways: feed or reels tab.

Try these questions on your own. I encourage to create some synthetic data on these tables, implement logic and share if you are comfortable. 

❓ SQL Question 1: For the last 7 days, find the number of unique users from US who watched 3 or more Reels via feed.

Output: day, users_count

❓SQL Question 2: For an experiment called "reels_comments_preview", compute for each variant (control vs treatment):
- Number of users with at least one watch each day, for 7 consecutive days, after the assignment date
- Completion rate = completed_watches / total_watches

Output: variant, users_count, completion_rate

Interviews are getting more practical. The best way to prepare is to simulate real product scenarios. To my mentee- Best of Luck :) 

Looking for any help or guidance on DS interview prep? Let's connect. 

#datascience #productDS #SQL

5 things a ML-focused data science resume must have.

I am not talking about the sections, but the content on it, mainly the bullet points. 

I have reviewed hundreds of DS resumes, and often times they are missing these key ingredients. 

🔹Production proof (not just notebooks): Deployed models, monitoring, retraining, latency, scaling. Show what you’ve shipped. If you are new grad, have atleast couple projects hosted in cloud with functional app for inference. It can be a simple app with an API , but far better than just code repo. 

🔹 Metrics that matter: Include model + business impact: AUC/PR lift, FPR reduction, cost saved, time saved, revenue impact. Often times business metrics are forgotten but thats the important one. Again, for new grads, try to use intern projects in business impact, if not then - articulate potential impact your project could create. Something is better than nothing. 

🔹 End-to-end ML pipeline ownership:
Data → features → training → eval → deployment → drift/alerts. Name the stack (Spark, Airflow, MLflow, Docker, K8s, etc.). Make sure the tools from your skills section get to justify where you used it. 

🔹 Real ML depth (not buzzwords): Call out problem type + approach: ranking, anomaly detection, NLP, time series, causal… and what you actually did (baselines, ablations, error analysis). Most of the times, people only mention problem types, without connecting it to approach. 

🔹Consistency in bullet points: answering what, how and why: Data Science is about storytelling and it starts with your resume. Make sure your bullets capture three things- what is done, how it is done, and what is the result/ impact.

Example: Built an anomaly detection model for account activity using Isolation Forest with rate/rarity features, improving precision 0.58 -> 0.76 and reducing false-positive investigations by ~35%, saving ~100 hrs analyst time weekly. 

#datascience #machinelearning #resume

taplio

Nirmal Budhathoki's Linkedin Analytics

Nirmal Budhathoki

Performance ScoreNirmal Budhathoki's LinkedIn Performance

What is Nirmal talking about?

Nirmal Budhathoki's Best Posts (last 30 days)

Ready to grow your LinkedIn presence?

taplio

.css-6ytdur{background-image:linear-gradient(284deg, rgba(39, 43, 72, 1) 0%, rgba(33, 47, 183, 1) 100%);color:transparent;-webkit-background-clip:text;background-clip:text;}Nirmal Budhathoki's Linkedin Analytics

Nirmal Budhathoki

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);}Nirmal Budhathoki's LinkedIn Performance

What is Nirmal talking about?

Nirmal Budhathoki's Best Posts .css-1emoo84{font-size:var(--chakra-fontSizes-md);font-weight:400;color:var(--chakra-colors-gray-600);}(last 30 days)

Ready to grow your LinkedIn presence?

Nirmal Budhathoki's Linkedin Analytics

Performance ScoreNirmal Budhathoki's LinkedIn Performance

Nirmal Budhathoki's Best Posts (last 30 days)