David

Reguera Suarez

Bibao, Basque Country, World

Hi, It's David, thanks for dropping by 👋🏼

Having been myself on both sides of the recruiting process, I know how dreary it can become. I’m decided, then, to minimize the drain for both of us. So, the moment anything just resonates, head over the CTA down there and drop me a line to book some time in to talk through the nitty-gritty details and save some precious time.

Now, let’s kick us off with the objective of all this business

Objective

I'm one of the lead developers on a fantastic team within a company with a very compelling mission.
However, it does not quite provide me with the opportunities of learning and growth I'm looking for.
So I'm actively seeking interesting projects that align with some long term view I devised
for myself while ago related to machine learning and data science.

In Short

A quick dive-in on what I can bring to the table

ML Ready

Ever since I attended to a TensorFlow seminar back in 2019, I've been sneaking in small ML projects here and there. And some others not that small like the one featured in the next section.

Code Craft

I take pride of my commitment to code cleanliness, whether it is slicing changes into meaningful commits or, iterating one more time to get that docstring spot-on.

Always Exploring

As you'll discover in my Elevator Pitch™️, curiosity is the main drive in my life. Whether it's learning CUDA, writing a neural network from the scratch or even installing Red Hat 6 on an old computer.

Timeline

We tend to overestimate what we can do in a day, and underestimate what we can achieve in a decade.
That’s why it’s sometimes desirable to have a big picture to see how everything fits in.

timeline-2024

A years scale like above implicitly assumes that years are equally productive, which is kind of misleading.
That’s the reason I added the equivalent STEM experience bit which, of course, carries its own caveats and modeling decisions.
Still, it provides a common framework to compare years one another and with a more general reality out there.

2023 Featured Project

Let's take a close look at last year's machine learning starred project .

Context

The company I’m working for nowadays is, at the end of the day, an insurance company and, as such, it needs to keep track of and review a fair amount of documents in order to fulfill the guarantees subscribed with their customers.
This is one of the bottlenecks in the business as it involves a significant amount of human intervention which makes the process very inefficient and prone to errors.
Therefore, this was the right place to implement some machine learning pipeline that could analyze and classify documents in an automated way.

Solution

As a side project to my main role in the company, I designed and implemented an end-to-end solution that is represented by the plot on the right. Click on the details to get more information.

  1. The main application upon upload from the user the relevant documents will store them on AWS S3 and call an inference endpoint, written in Flask and running on an EC2 instance, providing its url.  The endpoint will, then, acknowledge the request and run the inference as an async task.
  2. This task starts by sending the document url to AWS textract that will return the text contained in it.
  3. Once with the text, the inference endpoint will compute a pipeline of 5 steps with a variety of techniques: regex matching, document classification by means of the Spacy library, fuzzy search and the SuTime library.
  4. After these steps, it will return the outcomes of the inference back to the main application that will in turn store them in the DB.

Learning Experience

  1. If I learned anything, it was: be thorough and methodical in your experiments when you are figuring out the right model and hyperparameters.
  2. I ran through a couple tutorials: Kaggle’s computer vision (as in the beginning we were going to extract the text ourselves) & Huggingface NLP.
  3. The async task proved a challenge because of the NLP libraries
Inference Endpoint Sequence Chart
A mermaid representation of the sequence

Weak Spots

Not everything are bells and whistles.
So, it's often a good idea to be aware of oneself shortcomings so one can work on them.

No Formal Education

For better or worse I never believed in standard approaches to learning and eventually got along with the drawbacks they pose.

Overcoming this Downside

Whenever I learn something I try to be as honest as possible with myself always looking for the uncomfortable zone which is pretty much what an exam does. As Ericsson puts it, one needs 10k hours of deliberate practice.

Mathematical Uncertainty

While I love mathematics đť©­and find them really hard, by the wayđť©­ I'm unsure on whether my skills would be up to the task.

Overcoming this Downside

I have surveyed several forums and acquaintances and got several broad answers like oh, you need linear algebra or get into calculus, so one of my goals with these interviews is to figure this out more precisely.

Too Much Drive

I have a fair amount of drive which is a positive trait overall but that can lead sometimes to being less collaborative.

Overcoming this Downside

It’s always hard to mitigate a habit when one doesn’t know where is going to pop up. At most, one can aim for to be more and more aware to be prepared when the opportunity arises. This is exactly one of the pillars of my organization system.

Vanity Metrics

I'm a massive data nerd, a big fan of self quantification and being consistent in my endeavors
I can't imagine a better way to wrap up above sections than throwing some vanity metrics from 2023

0
Hours of STEM

It’s clear, by looking at the Equivalent STEM Experience, that a great deal of my annual time goes into STEM like activities, whether it’s coding, reading books or listening to podcasts.

0
Km Run

Does it seem like a lot? Don’t think so, it’s just a matter of consistently run 33k per week. Every week. Well, that, and cram in a 100-mile ultramarathon with all the training it requires.

0
Projects completed​

Does it seem like a few? Don’t think so, most of the people get very quickly into creating projects, the easy part, but very few manage to get them done. It’s a subtle art to tune the scoping though 

0
Hours Reading​

Which accounts for some 21 books featuring the inspiring The Science of Interstellar, the discovery of A Thousand Brains or the gem Chance and Necessity.

FAQs

If you have arrived so far and you are not still convinced, I can't help but show the FAQs

I’m committed to show the least thing that sparks interest with a curated list of things that I loved doing and am particularly proud of. So, if above bits haven’t quite worked, adding more stuff đť©­and, trust me, I have year detailed summaries all the way back to 2017đť©­ will have diminishing returns that won’t make this site any better.

And you are totally right. Linkedin solves the optimization problem of conveying the most information with the least amount of effort on the reader side as it provides a common framework to compare people one another. Because of my background, I won’t likely pass the minimum standards of that common framework so I have to come up with creative ideas to reach out a potential audience.

It’s a fair question. My stock answer is that for the last three years I’ve been having technical conversations with 100% formal-education-engineers on a daily basis and I’ve been up to the task. Put it another way, we could play a sort of computer machinery and intelligence and you may not be able to tell apart each of us. Still, one can make the case that there are levels of engineers out there, and very basic ones would be easy to match, so the first statement proves nothing.

Exactly what it sounds, a 5-minute sales pitch —I know, it should be two—  about my vision mission and values that will give you a glimpse on whether the collaboration will have a good product-market fit

Kind of, I must admit that I have been more of a maximizer than a satisficer but that is a trait that I started to change while ago. All in all, it’s a subtle art to find right balance time/quality.
Drop Me a Line