DEV Community: Nomfundo Mtiyane

Guide to use AI as a tool it is.

Nomfundo Mtiyane — Tue, 26 May 2026 04:42:29 +0000

AI Won’t Replace Developers.
But Developers Who Use AI Poorly Will Replace Their Own Bugs With Bigger Ones.

I keep seeing posts like:

“ AI wrote 10k lines of code that a developer will spend 2 years debugging .”

And honestly? That can happen.

Not because AI is “bad,” but because many developers are using it incorrectly.

AI is a power tool .
And power tools can build houses or cut through the floor.

So here’s a practical guide on how to use AI as a developer WITHOUT creating future technical debt, security risks, insane cloud bills, or impossible debugging sessions.

✅ USE AI FOR THIS

Boilerplate & Repetitive Tasks

Great use cases:

CRUD setup
React components
API routes
Form validation
Type definitions
Documentation
Unit test templates
SQL query drafts

2. Learning Faster

AI is incredible for:

explaining concepts
breaking down errors
comparing technologies
understanding architecture
debugging logic step-by-step

Use it like a mentor, not a magic machine.

3. Rapid Prototyping

Need to test an MVP idea quickly?

Perfect.

AI helps you:

ship prototypes
test UX ideas
validate business ideas
explore stack options

But prototypes are NOT production systems.

4. Refactoring

Good developers use AI to:

simplify messy code
improve naming
reduce duplication
improve readability
generate tests before refactors

❌ DON’T USE AI LIKE THIS

1. Blind Copy-Paste Development

If you paste AI code without understanding:

how state flows
why functions exist
where data comes from
how APIs interact

…you’re building future bugs.

If you can’t explain the code, you probably shouldn’t deploy it.

2. Generating Entire Architectures You Don’t Understand

A dangerous pattern:

“Build me a scalable microservices SaaS app.”

Now you suddenly have:

Redis
Kafka
Docker
Kubernetes
14 services
27 dependencies

…for a to-do app with 12 users.

Overengineering is now easier than ever.

Start simple.

3. Feeding Sensitive Information Into AI

This one is serious.

NEVER paste:

API keys
passwords
database credentials
private company code
customer data
authentication tokens
production configs
internal business documents

Some developers are literally pasting .env files into AI chats.

That’s dangerous.

Treat AI chats like public conversations unless your company has approved secure tooling.

4. Letting AI Choose Everything

Don’t ask:

“What stack should I use?”

without context.

AI can recommend:

trendy tools
unnecessary frameworks
expensive infrastructure
complex setups

Good engineering is about tradeoffs, not hype.

🔥 The REAL Skill In The AI Era

Before:

Developers were valued mostly for writing code fast.

Now:

Developers are valued for making GOOD technical decisions.

The most important skills today are:

system design
debugging
architecture
security awareness
performance thinking
communication
understanding business problems

AI speeds up coding.
It does NOT replace engineering judgment.

✅ Best Practices For Using AI As A Developer

Think in small chunks

Instead of:

“Build my entire app.”

Do:

“Help me design authentication.”
“Review this component.”
“Optimize this query.”
“Explain this error.”

Smaller prompts = better results.

Review EVERYTHING

AI can:

hallucinate functions
invent APIs
use deprecated syntax
create security flaws
introduce performance issues

Always verify.

Keep Architecture Simple

Simple systems:

scale better
debug easier
cost less
onboard developers faster

Complexity is expensive.

Test Constantly

Use:

logging
unit tests
integration tests
linting
type safety
monitoring

AI-generated code still needs engineering discipline.

Learn While Using AI

The best developers ask:

“Why?”
“What’s the tradeoff?”
“What problem does this solve?”
“Is there a simpler way?”

That mindset matters more than memorizing syntax now.

The Future of AI & Gaming in a Decentralized World

Nomfundo Mtiyane — Mon, 25 May 2026 16:12:35 +0000

Gaming is evolving beyond just “playing for fun.” We’re entering a world where AI, blockchain, and decentralized systems merge to create player-owned digital universes. The future isn’t just about better graphics anymore; it’s about intelligent worlds, autonomous economies, and giving players true ownership over their identities, assets, and experiences.

Traditional gaming platforms are centralized. Companies own the servers, economies, skins, characters, and even player data. In a decentralized gaming future, players become stakeholders. Items become tradable digital assets, AI agents become companions or competitors, and game worlds continue evolving even when developers stop updating them.

AI is becoming the brain of these worlds. Instead of repetitive NPCs following scripted dialogue, decentralized AI systems could create adaptive characters that learn from player behavior, evolve strategies, and participate in dynamic economies. Research into GameFi ecosystems already shows how embodied AI agents can create richer narratives, personalized gameplay, and smarter in-game financial systems.

One of the most exciting developments is the rise of fully on-chain games and autonomous worlds. These games live on blockchain infrastructure rather than private servers, making them permanent, transparent, and community-governed. Players won’t just play games ; they’ll help shape economies, governance, and even AI evolution inside those worlds.

Where OPUS AI Fits Into This Future

OPUS Layer AI is positioning itself as part of the infrastructure powering decentralized AI ecosystems. OPUS focuses on secure AI agents that can interact safely with blockchain systems, verify transactions, and automate complex digital operations.

In gaming, this could unlock major innovations:

AI-powered NPCs that evolve independently and adapt to player decisions.
Autonomous AI companions that can learn player preferences and strategies.
Dynamic in-game economies managed partially by AI agents.
AI-driven moderation and anti-cheat systems for competitive gaming.
Personalised quests, storylines, and matchmaking powered by decentralized AI.

For iGaming specifically, OPUS-style AI systems could become even more disruptive.

The Future of AI in iGaming

The online gambling and betting industry is already heavily data-driven. Combining decentralized AI with iGaming could reshape the industry through:

Smarter fraud detection and anti-money laundering systems.
AI-powered responsible gambling monitoring that detects harmful behavior patterns early.
betting experiences and adaptive game recommendations.
Provably fair decentralized casino systems using blockchain verification.
Autonomous AI dealers, tournament managers, and customer support agents.
Real-time risk management and odds optimization.

Because OPUS focuses on AI verification and secure AI-driven blockchain transactions, its infrastructure could potentially support safer and more transparent decentralized iGaming ecosystems where trust is built directly into the protocol layer.

The biggest shift, though, is ownership. In decentralized gaming, players may eventually own:

Their AI companions
Their gaming identities
Their assets and economies
Their reputation across multiple games

Instead of companies controlling everything, ecosystems become collaborative networks between developers, creators, players, and AI systems.

Of course, challenges still exist ; scalability, regulation, security, onboarding complexity, and balancing monetization with fun gameplay remain major hurdles. Community discussions around blockchain gaming show both excitement and skepticism about whether Web3 gaming can truly go mainstream.

Still, the convergence of AI + blockchain + gaming feels less like a trend and more like the foundation of the next digital era. The studios that succeed won’t just build games, they’ll build intelligent, decentralized worlds that players partially own and help evolve.

Why AI Won’t Replace Developers , But It Will Replace Lazy Engineering/vibe coders

Nomfundo Mtiyane — Mon, 18 May 2026 12:31:02 +0000

Published: May 18, 2026
Category: Engineering / AI / Software Industry

The hottest take in tech right now isn’t whether AI can code.

It’s whether developers still matter.

After spending the last year watching teams integrate AI copilots into production workflows, here’s the reality nobody on LinkedIn wants to say clearly:

“AI can generate code. It cannot generate accountability.”

That difference changes everything.

The Illusion of “AI Built My App”

Right now social media is flooded with posts like:

“Built a SaaS in 2 hours with AI”
“I replaced my dev team with agents”
“Vibe coding is the future”

Cool demos.

Terrible engineering strategy.

Because the moment the app:

scales,
gets attacked,
leaks data,
slows down,
corrupts transactions,
or crashes under concurrency...

…the human engineer suddenly matters again.

AI is excellent at generating possible code.

Senior engineers are responsible for generating:

reliable systems,
maintainable architecture,
observability,
security,
performance,
business continuity,
and technical judgment.

Those are not the same skill.

The Industry Is Quietly Shifting

The real shift happening in 2026 is this:

Junior developers who only learned syntax are struggling.

Developers who understand:

systems,
architecture,
product thinking,
UX,
cloud infrastructure,
and communication... are becoming more valuable.

The market is rewarding engineers who can:

- Think clearly,
- Verify AI output,
- Connect systems together,
- And make technical decisions under uncertainty.

That’s senior-level engineering.

What AI Actually Changes

AI removes:

repetitive setup,
boilerplate,
debugging friction,
documentation lookup,
and repetitive coding patterns.

Which means developers now spend more time:

designing,
reviewing,
architecting,
optimizing,
and shipping faster.

The best engineers today are becoming:

part developer, part product thinker, part systems designer.

That’s why full-stack development is exploding again.

Not because frontend + backend is trendy.

Because companies want engineers who understand the whole system.

- A Real Example

A developer can ask AI:

“Build authentication.”

And AI will absolutely generate authentication code.

But a senior engineer asks:

Which auth provider scales best?
How do we rotate tokens?
What’s our recovery flow?
What’s our threat model?
How do we prevent session hijacking?
What happens under rate-limited login attacks?
How does this affect onboarding conversion?

That difference is experience.

And experience compounds.

My Prediction for the Next 5 Years

The best developers won’t be the people who avoid AI.

They’ll be the people who:

direct AI effectively,
review critically,
architect intelligently,
and still understand fundamentals deeply.

AI is becoming a power tool.

And power tools don’t remove architects.

They just expose bad builders faster.

Final Thought

The future developer is not:

“just a coder”
or “just an AI prompt engineer.”

The future developer is a technical decision-maker.

Someone who can turn messy ideas into scalable systems.

That skill is still rare.

And rare skills survive technological shifts.

Task Tracking and Time Management for Productivity Optimization in Google Spreadsheets

Nomfundo Mtiyane — Mon, 18 May 2026 12:16:56 +0000

Overview
For anyone working in software development, managing daily tasks effectively is crucial. Without proper planning, managing time for complex tasks can become overwhelming. It may even lead to wasting time trying to organize unfinished work. That’s why a software developer must have precise time management skills.

While tools like Notion, Trello, Asana, ClickUp, and Jira are excellent for time and project management, Google Spreadsheets offers a minimalistic and straightforward alternative. In this blog, I’ll share two Google Spreadsheet templates: one for time management and another for tracking payment information. I’ll also explain how to use these templates and highlight their benefits.

Motivation
Software development and maintenance jobs come with numerous daily tasks. At the end of the day, reporting what you’ve worked on is often required. Without tracking your tasks, strategic planning becomes impossible.

When you start tracking tasks, you’ll notice where your time is going. Over time, you can identify unproductive tasks and eliminate them, allowing you to focus on your goals and targets. Using Google Spreadsheets, you can visually analyze your productivity through real-time charts, providing a bird's-eye view of your tasks. These insights make it easier to optimize your workflow and identify time-consuming activities.

When you open the Task Tracker Google Spreadsheet Click here to view it, the first thing you’ll see is a Column Chart.

The X-axis represents task types, and the Y-axis represents hours.
In the top-right corner, you’ll see the total number of tasks and hours for the day.
Next to the main chart is another chart displaying your daily hours, giving you an overview of how much time you spent on tasks each day.
Below the column chart is the Data Input Section, which includes four columns:

Date: The day you worked on the task.
Type: The type of task.
Task Name: A brief description of the task.
Time (hrs): The time spent on the task in hours.
Once you input data in these columns, the charts will automatically update to reflect your progress.

image

To make it easier, I’ve pre-defined some task types based on my job. These were identified with the help of data analysis.

Steps to Use:
Add a new row in the data input section.
Enter the task date and name.
Choose a task type from the dropdown menu.
Once the task is complete, strike it through to mark it as done.
Input the time spent on the task; the charts will update instantly.
This system helps you monitor your completed tasks and evaluate how your time is distributed.

In addition, I’m sharing another Google Spreadsheet designed for tracking university tuition fees.

Advantages
Simplified Task Tracking: Easily input and manage daily tasks without complex tools.
Real-Time Visualization: Automatically updated charts provide instant insights.
Efficiency Monitoring: Quickly identify time-consuming tasks and optimize your workflow.
Customizable: Tailor the spreadsheet to suit your specific needs and job type.
Cost-Effective: Utilize a free tool for robust task and time management.
Conclusion
Google Spreadsheets is a versatile tool for task tracking and time management, especially if you prefer a minimalistic and efficient solution. With its real-time charts and customizable data inputs, it’s perfect for software developers and professionals looking to streamline their daily workflow.

By regularly tracking your tasks and time, you can make informed decisions, improve focus, and ultimately increase productivity. Explore the templates shared in this blog and start optimizing your productivity today!

How much can a Front-end Developer learn about Machine Learning using only JavaScript?

Nomfundo Mtiyane — Wed, 06 May 2026 10:22:57 +0000

Robot Playing Piano by Franck V on Unsplash: https://unsplash.com/photos/U3sOwViXhkY
Machine Learning and Artificial Intelligence have been huge buzzwords in the Tech industry for quite some time. Hearing them might make you picture self-driving cars or chat bots in your head and leave you wondering what kind of programming goes into those projects. AI, in particular, has been the stuff of science fiction for so long that it’s hard not to be curious about it now that we’re seeing real-life projects surface in the news.

But what do those terms really mean? A cursory search of any of them will undoubtedly leave you separating science from marketing before you get to any valuable information. I’ve spent the past few months learning and researching Machine Learning and the Data Science field, so I know how difficult it can be to drill through the buzz and hype. This article is about that experience and how far I was able to get as a humble JavaScript developer.

Before we get started though, I want to kick things off with my favorite quote about Machine Learning and AI:

Difference between machine learning and AI:

If it is written in Python, it’s probably machine learning

If it is written in PowerPoint, it’s probably AI

—Matt Velloso on Twitter

I thought Machine Learning was only for geniuses (and I’m no genius)
I first started working with Machine Learning (ML) early on in my career when working with a team that was doing text-recognition work. My job was to create a front-end application that let users send documents to the backend for parsing.

Python is the main squeeze when it comes to ML/AI work for a lot of reasons, so I thought I wouldn’t get much exposure to any ML work as a front-end developer. However, I ended up tangling with ML quite a bit. Text Recognition models are pretty picky about input, so I ended up using JavaScript implementations of Python libraries to pre-process documents in the browser to match the backend. Through this, I also found JavaScript versions of Machine Learning frameworks as well. Not only could you use trained models to predict data in the browser, but you could create and train models there too.

While it was awesome to see that JavaScript can really do all the things, Machine Learning is a specialized field that has massive foundations in calculus and statistics. So while I felt competent in my JavaScript abilities, I felt incompetent when it comes to the kind of higher level math involved in ML. In short, I thought the whole data science field was beyond me as a lowly front-end developer.

After moving on from that work project, I didn’t touch Machine Learning for months.

You don’t need to be a Mathematician to try your hand at Machine Learning
Pic of Enrico Fermi doing some math. May or may not be ML-related. Looks hard. Photo by Science in HD on https://unsplash.com/photos/aYxQrt5J6jM
(Pictured: What I thought all Data Scientists looked like before this journey.)

Even though I gave up on ML, I still wanted to learn Python. I run a meetup, so I decided to get some workshops together to teach the community (and myself) Python. I reached out to the local Python User Group. Its organizer, Michael DuPont, and I got together and planned out a series of talks and workshops where Michael would teach, and I would book the venues.

The workshops were awesome, and we were also able to record them for YouTube. That experience taught me a lot about Python and its ecosystem, but also some interesting lessons about Machine Learning.

At the end of the intro talk, Michael introduced TPOT—an automated ML tool—and used it to train a model to predict housing prices in Boston. This was fascinating to me because what TPOT does is take data and figure out how to make a model and train it on its own. It essentially cut out all the math and statistics for you. This planted the seed in my head that maybe, just maybe, being a developer is enough to get your feet wet with ML.

Using JavaScript for Machine Learning
I also run a podcast, and through that, I ended up meeting Gant Laborde, a JavaScript developer with a passion for data science. We interviewed Gant about Machine Learning in JavaScript, and really dug into how to get started with ML.

Gant also revealed that he was working on a JavaScript Machine Learning course. This piqued my interest because almost all ML tutorials focus exclusively on Python. After we were finished recording, I offered myself up as a guinea pig for Gant’s course, and he graciously sent my co-host and I copies of the course to try out ourselves.

Shortly after, Gant released a free intro to Machine Learning course. Naturally, I signed up for that too.

The main course is meant to take 3 weeks, and the intro course is meant to be a 5-day course.

…It took me about 3 months to work through both courses.

My completion time ballooned past the projected 4 weeks for both courses because of my busy schedule, but also because I truly wanted to grasp the material and give myself every chance to learn everything that Gant had poured into the course.

So I took my time ⏳, drank many cups of coffee ☕, and wrote a lot of code 👨‍💻.

Let’s talk about what I’ve learned 🧙‍♂️.

Machine Learning is really about prediction 🔮
The general intro course was a fantastic intro to Machine learning. It taught me that ML is really about boiling a data set down to numbers, analyzing a huge group of those numbers, and then being able to predict outcomes when given data it hasn’t seen before. It also taught me about the types of Machine Learning and their applications in real life.

The Machine Learning and Data Science that we deal with are more about Artificial Narrow Intelligence (ANI) than Artificial General Intelligence (AGN). AGN is the stuff of science fiction: robots that can function like humans and make decisions for themselves, SkyNet, etc. ANI is about focusing on a specific problem or question. Some examples of ANI include figuring out if a picture is of a cat or a dog, predicting sales numbers based on historical data, or detecting when a person’s eyes are open from a video feed.

But how does an algorithm learn to answer those questions?

In the example of the Boston Housing data, the data set has a bunch of features like location, crime rate, proximity to schools, etc. that the computer analyzes alongside the price of the home. An ML algorithm will read thousands of those data points to be able to approximate a home price based on those features.

Think about a line from your middle school math classes. There’s a formula that will tell you what x and y coordinates will fall on that line with 100% accuracy. Let’s say our line has a formula of y = x. Using that formula, we could very easily figure out if a set of coordinates are on that line, right? If you have points (0, 0), (1, 1) and (2, 2), you know that as long as they’re equal, they live on the line.

But how would Machine Learning approach this problem? Imagine you didn’t have a formula that could tell you with 100% accuracy whether a point was on the line or not. How could ML help?

To solve this with ML, you would feed your model thousands of coordinates that are labeled as on or off the specified line. After doing a lot of math and burning a lot of processing time, you’d have a model that could tell you with a certain confidence (a percentage) how likely any given point is to exist on that line.

In a nutshell, Machine Learning is just a computer clumsily learning through trial and error.

This joke tweet really encapsulates the core concept of Machine Learning:

Machine Learning Job Interview:

Me: I’m an expert in machine learning

Interviewer: What’s 9 + 10?

Me: It’s 3.

Interviewer: Not even close. It’s 19.

Me: It’s 16.

Interviewer: Wrong. Its still 19

Me: It’s 18.

Interviewer: No, it’s 19.

Me: It’s 19.

Interviewer: You’re hired.

What kind of coding is involved in Machine Learning?
Thankfully, the inner-workings (calculus and linear algebra 💀) of training a model are abstracted away from us by ML frameworks like TensorFlow. So we don’t have to construct the actual algorithms used to process data and train models.

However, there’s still a level of math that you have to grapple with when dabbling in Machine Learning. You need to first be able to process data to pass into ML algorithms and models. You also need to have some knowledge of ML framework settings and configuration.

Preparing Data
Most of the work done by data scientists is involved in preparing the data. When we interviewed Data Scientist Amelia Bennett on our podcast, she described herself as a high-paid data janitor and described data science itself as a “21st century dirty job”.

If you’ve ever wondered how a computer can learn to parse images, sounds, and language, the answer is math. Anything that can be described mathematically can be translated to numbers and fed into ML models. The job of the data scientist is to not only select the data, but convert it. In computer vision for example, this means converting images to arrays of pixels (RGB and location) that the algorithm can use for training. Natural language processing involves describing soundwaves using math—taking frequency and pitch numbers over time to identify spoken words.

Fortunately, there are tools to help you convert non-number data to numbers out there. Tensorflow has a lot of utilities dedicated to helping you process images for instance. Still, you need to be able to use those tools and know which ones to reach for.

Training Configuration
An audio mixer with a mess of cables and knobs by Steve Harvey on https://unsplash.com/photos/xWiXi6wRLGo
Machine Learning frameworks abstract away most of the math, but you’re still stuck needing to know how different algorithms affect training. So you won’t be solving any problems on paper, but you’re still going to be reading about mathematical concepts with scary names like softmax, sigmoid, and ReLu.

There’s also something to be said about the amount of terms like those activation function names I mentioned above. When using tensorflow specifically, you might feel like you’ve got more knobs to twist and turn than you really know what to do with. The TensorFlow API is massive, and it’s hard to imagine someone mastering all of it.

Note: You may remember that there are autoML frameworks like TPOT that I mentioned above. These don’t require a ton of configuration, but also give you less control over the outputted result. While these tools are useful, you’ll likely find yourself using something TensorFlow or something similar the further you get into ML.

Guess Work
I personally found the amount of configuration at my fingertips to be completely overwhelming. I am the type of person that loves to know exactly what I’m doing and why. However, this left me at odds with a data scientist’s typical workflow.

Machine Learning requires a lot of experimentation. I used to think that data scientists trained models in one go, but in reality they may train models over and over again before getting desirable results. When training models, there’s many various settings to tweak, and selecting the right ones is more of a matter of trial and error than anything else. I had to let go of my need to understand and comprehend everything before I could embrace the experimentation required to solve ML problems.

To put it another way, data scientists are a bit like fictional mad scientists haphazardly mixing chemicals in a lab—except data scientists are mixing mathematical functions together instead of fluid-filled beakers of various colors.

Examples of Machine Learning with JavaScript
I got a solid foundation in the basics of how Machine Learning works, but what projects have I been able to build with it? Before we get to the code, let’s talk about how to employ ML as a developer.

There’s really two main types of projects when you’re working with Machine Learning: using a pre-trained model, or building and training your own model. I’ve included examples of both below.

Note: These demos were all made from concepts and lessons learned from Gant Laborde’s AI course.

Pre-trained Model Examples:
These are super fun apps to make, and require almost no ML knowledge to pull off. I recommend trying some of these yourself! You can quickly make some impressive ML projects by utilizing browser APIs and web technology.

Image API + MobileNet Example
Gif displaying mobilenet classifying random photos
In this demo, I pulled images from the lorem picsum API and used React to hook them up to MobileNet, which is a pre-trained model that can classify images.

Lorem Picsum is mostly random artsy still-life and landscape photos from Unsplash, so you get some interesting results from MobileNet’s classifications.

Webcam + MobileNet Example
Gif of me getting mobilenet to recognize various objects
This app connects MobileNet up to the webcam browser API so that you can point your phone/camera at an object and classify it.

(If you’re wondering about some of the crazy results, you should know that MobileNet wasn’t trained recognize humans.)

Browser-based Model Training Examples:
Now we get into the heavy stuff. Beware running these on low-end phones and devices.

These examples take data sets and use them to train models directly in your web browser using JavaScript!

Solving FizzBuzz with TensorFlow
Gif showing a model finishing training and solving fizzbuzz
This example feeds a model thousands of numbers (100 through 3100) that have been solved (labeled) using a simple fizzbuzz algorithm. Then it tries to guess whether numbers 1-100 should be fizz, buzz, or fizzbuzz.

Cat/Dog image recognition with TensorFlow
Gif of my dogs-n-cats image recognition demo app recognizing random dogs and cats
This app uses Gant Laborde’s dogs-n-cats npm package to train a model to recognize random dogs or cats.

Click here to check out the demo on codesandbox. Be warned though, it’s a bit memory/resource intensive to train a model using 2000 images in the browser.

The dogs-n-cats package does most of the prep work for you behind the scenes by pre-processing and pre-packaging all 2000 dog/cat images into tensors for you. All I had to do was feed the images directly into a model for training.

Other Types of Machine Learning
The examples above all involve supervised learning, meaning we are telling the algorithm what to look for and giving it labeled examples to learn from. Supervised learning is fairly simple to wrap your head around, but there’s more methods and applications of ML out there.

Two examples of different ML methods are unsupervised and reinforcement learning. In unsupervised learning, you give the algorithm a data set that’s unlabeled and let it discover and classify things on its own. With reinforcement learning, where the algorithm learns how to accomplish tasks through good or bad outcomes. Think of a computer learning to beat a mario level as an example.

I’d love to dip my toes into these other types of ML at some point, but believe me when I say that getting this far with supervised learning was a huge milestone all on its own!

Are you a Data Scientist now?
Photo of the word "nope" in cursive by Daniel Herron on https://unsplash.com/photos/vBxbZokRL10
Let’s get this out of the way: I am definitely not a data scientist after spending a little time with Machine Learning.

That said, I have learned a lot about what is actually involved in Machine Learning and how Data Scientists have been able to pull off some of the incredible advances we’ve seen in the field. I have a solid understanding of the magic (read: math) that’s behind computer vision, natural language processing, and other miracle technologies.

I also understand the limitations of Machine Learning and know that there’s still quite a lot of work to be done in the field. I’m excited to see what happens as more data sets and better, more accessible ML frameworks become ubiquitous.

As far as training models, I definitely feel like I’m still a novice. There’s so much nuance and intuition involved in selecting proper activation functions, filters, epochs, etc. that I hardly feel qualified to do anything more than guess at how to properly train a model.

I’m also well aware of the massive shortcuts in data preparation I employed. I feel confident that I could train a model from spreadsheet data, but when it comes to preparing and labeling images for training, I still have a long way to go.

Parting Thoughts
Data Science and Machine Learning are really cool things to dip your toes into if you can stomach the mathematics and trial-and-error that’s involved. It’s a challenging area of technology and something that businesses have begun to heavily invest in, so even having a basic understanding of the data science field is invaluable as a developer.

I definitely recommend Gant’s free intro course to ML/AI concepts. If you complete that course and feel like you want to pursue the subject further, I recommend checking out Gant’s paid beginner course on AI/ML in JavaScript as well. Gant worked very hard to make the concepts within approachable and it shows.

I hope to keep advancing my own Machine Learning skills in the future. If you have ideas on where I should go from here, or questions about this post, hit me up on Twitter! I’d love to hear from you.

SpaniLink?? What did I build

Nomfundo Mtiyane — Wed, 06 May 2026 10:19:45 +0000

Job hunting is exhausting.

Not just because of rejection , but because of how messy the entire process is.

At one point, I was applying to so many roles that I completely lost track.
Which company did I apply to?
Did they respond?
Was I rejected, ghosted, or still under consideration?

Everything started to blur.

And the frustrating part? I was actually getting noticed.

I made it to final-stage interviews. I’ve built projects, joined hackathons, and put in the work. From the outside, it looked like progress.

But behind the scenes, my job search felt disorganized, repetitive, and honestly… discouraging.

That’s when it clicked:

Job searching isn’t just about effort — it’s about systems.

The Problem No One Talks About

Most advice focuses on:

Fix your CV
Build projects
Practice interviews

But no one talks about managing the process itself.

Your applications end up scattered across:

Emails
LinkedIn
Company websites
Spreadsheets

There’s no single place that gives you clarity.

No system that tells you:

How many applications you’ve sent
Which ones are progressing
Where you’re getting stuck
What you should improve

So I decided to build one.

What is SpaniLink?

SpaniLink is a smart job application tracking platform designed to help job seekers take control of their job search using data and AI.

Instead of juggling spreadsheets, emails, and scattered notes, SpaniLink brings everything into one place — your applications, statuses, and insights — so you always know where you stand.

At its core, SpaniLink solves a real problem:

Job searching is messy, repetitive, and often discouraging.

Most job trackers today simply store information.

SpaniLink goes further.

💡 What Makes SpaniLink Different?

SpaniLink is not just a tracker — it’s a decision-making and automation system.

It’s designed to:

Automatically track job applications from emails
Organize applications into clear stages (Applied → Interview → Offer)
Provide insights on your job search performance
Reduce manual work and email clutter

Think of it as your personal job search assistant, not just a database.

🔥 What I’m Building Next

SpaniLink is evolving into something much bigger. Here are two powerful features I’m currently working on:

✍🏽 AI CV Tailoring

One of the biggest reasons people don’t get interviews is because their CV isn’t tailored to the job.

SpaniLink will allow you to:

Paste a job description
Automatically tailor your CV to match the role
Highlight relevant skills and keywords
Improve ATS compatibility

👉 The goal:
Stop sending the same CV everywhere , start sending the right CV.

🛡️ Email Phishing & Scam Detection

Job seekers are increasingly targeted by scams and fake job offers.

SpaniLink will:

Scan incoming job-related emails
Detect phishing attempts and suspicious patterns
Flag fake recruiters or scam offers
Help protect users from fraud

👉 The goal:
Make job searching safer , especially for students and graduates.

🌍 The Vision

SpaniLink is being built for:

Students
Graduates
Job seekers navigating competitive markets

The long-term goal is to create a platform that:

Tracks your journey
Learns from your data
Gives you personalized advice
Helps you land a job faster

What I’ve Learned So Far

Even while building this, I’ve realized:

Being qualified isn’t always enough
Visibility matters just as much as skill
Consistency beats motivation
And most importantly , effort without structure leads to burnout

Final Thought

Job searching shouldn’t feel like shouting into the void.

With SpaniLink, the goal is simple:

Turn your job search into a structured, intelligent, and data-driven process.

I’ll be sharing my journey as I build this , the wins, the struggles, and everything in between.

If you’re also trying to break into tech or navigate the job market, you’re not alone. 🚀