The real difference between engineers and non-engineers ‘vibe coding’. Part 2 of a three-part series on AI, engineering, and what actually changed.

If AI can write the code, what’s an engineer for? Part 1 of a series on AI, engineering, and what actually changed.

I’m rebooting this blog to bridge the gap between deep systems engineering and technical leadership

In this blog post I will explore the steps to optimise a sparse vector dot product operation. We will start with a basic implementation in Go, convert it to assembler and then iteratively optimise it, measuring the effect of each change to check our progress. All the code from the post is available on Github and also forms part of the Golang Sparse matrix package. A vector dot product is a very common kernel, or basic building block, for many computations used within scientific computing and machine learning e.g. matrix multiplication, cosine similarity, etc. The dot product simply multiplies together the respective elements of two vectors and then adds all the results together to give a scalar output. ...

This is the third in a series of blog posts sharing my experiences working with algorithms and data structures for machine learning. These experiences were gained whilst building out the nlp project for LSA (Latent Semantic Analysis) of text documents. In Part 2 of this series, I explored sparse matrix formats as a set of data structures for more efficiently storing and manipulating sparsely populated matrices (matrices where most elements contain zero values). We tested the impact of using sparse formats, over the originally implemented dense matrix formats, using Go’s inbuilt benchmark functionality and found that our optimisations led to a reduction in memory consumption and processing time from 1 GB to 150 MB and 3.3 seconds to 1.2 seconds respectively. ...

This is the second in a series of blog posts sharing my experiences working with algorithms and data structures for machine learning. These experiences were gained whilst building out the nlp project for LSA (Latent Semantic Analysis) of text documents. In Part 1 of this series, I explored alternative approaches for representing and applying TF-IDF transforms for weighting term frequencies across document corpora. We tested the approaches using Go’s inbuilt benchmark functionality and found that our optimisations materially improved not just memory consumption but also performance (reducing memory consumption and processing time from 7 GB and 41 seconds to 250 KB and 0.8 seconds respectively). In this blog post I shall explore other areas for optimisation, seeking to further reduce memory consumption and processing time. ...

In my last blog post I walked through the use of machine learning algorithms in Golang to analyse the latent semantic meaning of documents. These algorithms, like many others in data science, rely on linear algebra and vector space analysis. By their nature, they often have to deal with large data sets, so any inefficiencies in the data structures used or algorithms themselves can result in a large impact on overall performance and/or memory usage. Inefficiencies that are negligable when working with small data sets can have a huge cost applied across extremely large datasets. As memory is a constrained resource, this could end up limiting the size of data sets that may be processed (certainly without having to resort to persistent storage and/or alternative algorithms) or the types of algorithms used. To this end, I decided to see if I could optimise the algorithms I used to consume less memory and improve processing performance without sacrificing too much functionality or accuracy. This is the first in a series of articles sharing my experiences benchmarking and optimising the algorithms and data structures used whilst building out the nlp project. ...

I spend a lot of time reading articles on the internet and started wondering whether I could develop software to automatically discover and recommend articles relevant to my interests. There are various aspects to this problem but I have decided to concentrate first on the core part of the problem: the analysis and classification of the articles. To illustrate the problem, lets consider the following string representing an article for the purpose of this example. ...

A little over a year ago, I wrote a blog post examining Socratic Questions. Socratic Questions are a method of pull influencing that can be used to stimulate critical thinking. To help make the question types easier to understand and remember for use in practice, I have gone back and created an infographic illustrating the 6 types of questions. The infographic is shown below (click on the infographic for the full size version). ...