Toptal

Machine Learning engineer / Data Scientist

Braintrust

Machine Learning Engineer | Founding Talent Member

EasyHealth

Machine Learning Engineer

Cogsy

Time series ML engineer

Speedy Mentors

Mentor

Fifteenkay

Machine Learning Consultant

Speak

Data Engineer

Lola Market

- Developed, deployed, and maintained a Machine Learning model to improve the efficiency of the shoppers' fleet.
- Bootstrapped the first data-warehouse and reporting layer in the company (Amazon Redshift + Amazon DMS + Tableau).

Iridium Advisors

- Statistical analysis of financial markets using explainable Machine Learning techniques.
- Benchmarked a range of ML solutions to enhance model accuracy and explainability.
- Wrote a Python package to ensure the in-house reproducibility of each step of the analysis (data processing, data validation, data visualization, model construction, model validation, and model explanation).

cyngn

- Developed and maintained the analytics dashboards for Cyngn's engineering team and Cyngn's clients.
- Developed components of the internal ETL tool in Python and SQL.
- Assisted the back-end engineer to integrate both frontend and backend of the analytics stack inside Amazon Redshift.

Data scientist

Author of the Hands-on Reinforcement Learning course 🚀

Data scientist and ML engineer with over 8 years of experience. I have a passion for building ML-based solutions, from development to deployment. I love transforming an idea into a model, and a model into an API/product.

I have worked on different problems: financial derivative pricing, digital marketing analytics, deep learning for art generation, or demand prediction for on-line shopping. I have a background in pure mathematics and strong coding skills in Python.

The Mobile Small Language Model Revolution is here ✊✊🏾✊🏻
Here’s an example ⬇️
  
  
𝗪𝗵𝗮𝘁 𝘁𝗵𝗲 𝗵𝗲𝗰𝗸 𝗮𝗺 𝗜 𝘁𝗮𝗹𝗸𝗶𝗻𝗴 𝗮𝗯𝗼𝘂𝘁? 🤔

After years of seeing LLMs confined to cloud APIs and hefty infrastructure, I decided to prove a point: Small Language Models can run efficiently on mobile devices.

I truly believe the future (and present) of LLM engineering is not about integrating your application or service with a third-party-API-all-in-one-chatbot (aka GPT-5 and friends), but building (aka fine-tuning) and locally deploying a small task and device specific LLM for your business case.

Because accurate chat completions are cool, but having them

> Faster (less than 300ms)
> Cheaper
> Without sharing your company (and customer) data

is KEY if you want to stay ahead of the competition.


𝗛𝗮𝗻𝗱𝘀-𝗼𝗻 𝗲𝘅𝗮𝗺𝗽𝗹𝗲 👨🏻‍💻

My latest open-source project is a chess iOS application that embeds a fine-tuned SLMs directly on the device

↳🔗 https://lnkd.in/egvCFjtA

Here's what makes it special:


𝗞𝗲𝘆 𝗧𝗲𝗰𝗵𝗻𝗶𝗰𝗮𝗹 𝗛𝗶𝗴𝗵𝗹𝗶𝗴𝗵𝘁𝘀 🎯

1️⃣ Open-weight small Language models fine-tuned on a custom instruction dataset.
I used Liquid AI s powerful small models (more precisely the 350M version), and a custom dataset of chess moves I built using Chess Grand Master moves.

2️⃣  On-device inference using the Leap Edge SDK.
No API calls, no internet required, no data shared with a third-party.

3️⃣ Resource optimization
Runs smoothly on iOS hardware constraints.

4️⃣ Real-time interaction
Sub-second response times (less than 200ms)


𝗧𝗵𝗲 𝗯𝗶𝗴𝗴𝗲𝗿 𝗽𝗶𝗰𝘁𝘂𝗿𝗲 🖼️

This isn't just about chess. It's about proving that AI can be:

✅ Privacy-first (data never leaves the device)
✅ Cost-effective (no cloud compute costs)
✅ Accessible (works offline)
✅ Performant (optimized for mobile)

For LLM engineers tired of cloud-only solutions, this is your blueprint for bringing intelligence to the edge.


What would you build if you could run LLMs natively on mobile?á
Please let me know on the comments 👇

----
Hi there, I am Pau!

Glad you got to the end of the post (who reads these days?!)

I share tips and tricks about building AI products that work. If you want to get more of these posts and have a higher signal-to-noise ratio in your feed, 𝗴𝗶𝘃𝗲 𝗺𝗲 𝗮 𝗳𝗼𝗹𝗹𝗼𝘄 Pau Labarta Bajo!

#LLM #MobileAI #SmallLanguageModels #iOS #EdgeAI #MachineLearning

Pau Labarta Bajo

Building AI systems that work > Maths Olympian > Father of 1.. sorry 2 kids

How do I deploy a Language Model to mobile? 🚀📱
Here's the complete engineering workflow ⬇️
  
  
𝗪𝗵𝗮𝘁’𝘀 𝘁𝗵𝗲 𝗽𝗿𝗼𝗯𝗹𝗲𝗺? 🤔
Every LLM engineer knows the theory: fine-tune a model, optimize it, deploy it, and blablabla.

But when it comes to mobile deployment, most engineers hit a wall between "model trained" and "app published."

I just shipped a complete training-to-iOS-App pipeline that any LLM engineer can follow.

No hand-waving.
No "left as an exercise."
Just production-ready code using open-weight models and open-source tools by Liquid AI


𝗛𝗮𝗻𝗱𝘀-𝗼𝗻 𝗲𝘅𝗮𝗺𝗽𝗹𝗲 👨🏻‍💻

The Github repository is here
↳🔗 https://lnkd.in/egvCFjtA

Here's the exact engineering workflow I used:


𝗦𝘁𝗲𝗽-𝗯𝘆-𝗦𝘁𝗲𝗽 𝗜𝗺𝗽𝗹𝗲𝗺𝗲𝗻𝘁𝗮𝘁𝗶𝗼𝗻 🛠️

Step 1️⃣ Data Pipeline
Automated download of 7K chess games in PGN format from historical databases.

Step 2️⃣ Dataset Engineering
Extract game states, valid moves, and expert decisions into supervised fine-tuning format.

Step 3️⃣ Fine-tuning with Modal
Used Modal for serverless GPU access + Unsloth AI for 4x faster training. Only 5K steps needed!

Step 4️⃣ Model Evaluation
Critical step: measuring performance on out-of-distribution scenarios (because chess is complex!).

Step 5️⃣ Mobile Model Bundling
LoRA weights + base model → single .bundle file using the Leap Edge SDK CLI by Liquid AI

Step 6️⃣ iOS Integration
Complete Xcode project with LeapSDK, ChessKit, and ChessboardKit dependencies.


𝗧𝗵𝗲 𝗯𝗶𝗴𝗴𝗲𝗿 𝗽𝗶𝗰𝘁𝘂𝗿𝗲 🖼️

This pipeline works for ANY domain where you need:

✅ Custom expertise (legal, medical, financial advice)
✅ Privacy requirements (healthcare, enterprise data)
✅ Offline functionality (field work, remote areas) 
✅ Cost optimization (high-frequency inference)

For LLM engineers ready to move beyond POCs and ship production mobile AI, this is your complete playbook.

What domain-specific mobile LLM would you build using this workflow?
Please let me know in the comments 👇

-----
Hi there, I am Pau!

Glad you got to the end of the post (WHO reads these days?!)

I share tips and tricks about building AI products that work. If you want to get more of these posts and have a higher signal-to-noise ratio in your feed, 𝗴𝗶𝘃𝗲 𝗺𝗲 𝗮 𝗳𝗼𝗹𝗹𝗼𝘄 Pau Labarta Bajo!

#MLOps #MobileMLOps #ProductionML #LLMEngineering #iOS #EdgeComputing

Technical interviews aren't testing what you think they are.
Here is my advice ⬇️
  
  
𝗪𝗵𝗮𝘁'𝘀 𝘁𝗵𝗲 𝗽𝗿𝗼𝗯𝗹𝗲𝗺?

Most candidates obsess over memorizing every algorithm, paper, and implementation detail.

They walk into interviews trying to prove they know everything about the topic. 
But this approach misses the point entirely.

What you are being tested at this stage is not whether you know everything about the topic.


𝗧𝗵𝗲 𝘀𝗼𝗹𝘂𝘁𝗶𝗼𝗻

Understand what interviewers actually evaluate:
• 𝗧𝗲𝗰𝗵𝗻𝗶𝗰𝗮𝗹 𝗹𝗶𝘁𝗲𝗿𝗮𝗰𝘆: Can you discuss concepts accurately and connect related ideas?
• 𝗟𝗲𝗮𝗿𝗻𝗶𝗻𝗴 𝗮𝗯𝗶𝗹𝗶𝘁𝘆: How do you approach unfamiliar problems? Can you reason through solutions?
• 𝗠𝗶𝗻𝗱𝘀𝗲𝘁: The capacity to reason, learn, zoom in, zoom out and synthetize is even more important than knowledge.
• 𝗥𝗮𝗺𝗽-𝘂𝗽 potential: Will you be productive quickly? Can you adapt and grow?

Smart companies hire for skills over knowledge because:
> Knowledge becomes outdated quickly (especially in AI)
> Skills transfer across problems and domains
> Reasoning ability matters more than memorization

Next time you prep, focus less on cramming facts and more on developing your problem-solving approach.

Show them how you think, not just what you know.

Companies (at least the smart ones) hire more for skills than for knowledge.

---
This is just an extract from my latest post at the Real World ML Newsletter.
You can find the link below on the comments ⬇️⬇️⬇️

Taplio

Pau Labarta Bajo's Linkedin Analytics

What is Pau talking about?

's Best Posts (last 30 days)

Want to drive more opportunities from LinkedIn?

Taplio

.css-1jgreyo{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;}Pau Labarta Bajo's Linkedin Analytics

What is Pau talking about?

's Best Posts .css-qqfgvy{font-size:var(--chakra-fontSizes-md);}(last 30 days)

Want to drive more opportunities from LinkedIn?

Pau Labarta Bajo's Linkedin Analytics

's Best Posts (last 30 days)