Preach what you practice

Turning caffeine into code and bugs into features. Join us as we code, create, and innovate - one line at a time!

Has the AI singularity already begun? Where AI stands today and what could come next

AI assistants now answer questions, while agents can plan, use tools and complete substantial work. That shift can feel like the beginning of something much larger. I examine how close today’s systems are to the AI singularity and what such a transition could mean for work, healthcare, education and everyday life.

Our AI workflow, from Jira ticket to pull request

We build solutions for enterprises whose products reach more than 20 million people, which means most of our work is under NDA, which makes it tricky to work using third party tools, especially AI. What follows is our setup written down as a playbook you can copy, with client data privacy in mind and tips and tricks that helped us the most.

Using Claude as a design partner: why the most useful AI tool in your workflow isn't generating anything

AI in design is mostly discussed in terms of output - screens generated, wireframes produced, assets created faster. This is about a different use: Claude as a thinking partner before anything gets built. Pressure-testing information architecture and reframing what user research is actually saying. Not generation. Validation. Below are two prompts we actually use, what came back, and where the approach cost us time rather than saving it.

AI in QA: partnership or rivalry

Artificial intelligence is reshaping quality assurance by automating repetitive tasks and helping teams deliver software faster than ever before. While some people worry that AI will take over QA jobs, I see it as a powerful partner that enhances our productivity and efficiency. The real question is whether AI should be seen as a rival or a partner, and why human creativity remains irreplaceable in the testing process. The future of QA isn't about robots replacing us, it’s about humans and AI working together.

World Cup 2026: building football analytics with open-source computer vision

Imagine building your own football analytics engine, one that watches a match and rebuilds it as a live top-down minimap, draws a ring around every player and tracks their movement, marks whoever has the ball, and keeps a running count of possession, passes and turnovers, all from an ordinary broadcast clip and nothing else.

AI for QA: keeping your product private

Working across different QA projects has taught me that a useful tool is not always the tool with the most access. AI can speed up test case writing, API test preparation and automation drafting, but it does not need to connect directly to the product, repository or private environment. I used AI with controlled inputs such as feature flow videos, HAR files, screenshots and repo structure examples. That gave me useful output while keeping the application, sensitive data and VPN-protected environment private.

Setting up Claude Dispatch for Design QA

There are better ways to spend a Thursday afternoon than playing spot the difference with your own designs. This is a step-by-step guide to setting up Claude Dispatch for Design QA, giving it access to your environment and Figma so it can do the comparing for you.

AI tokens: A new default for open source contribution?

A few weeks ago an issue surfaced on the Gas Town repository alleging that two of its default formulas quietly contribute upstream PRs using users' Claude credits and GitHub accounts, with no opt-in and no README mention. The reported behaviour was never made opt-in, and the proposed fix was closed by maintainers. Looking past the specific case, the underlying mechanism could become a useful default for open source, if it were structured as a token tithe with explicit consent, scope, and attribution. This post sketches what such a tithe might look like, how it could plug into existing tooling, and why the loop it creates might address problems that traditional sponsorship has not.

AI in Project Management - Threat or opportunity?

AI is now a standard part of the job description across many industries and in the software world it has been omnipresent for a while. As a Project Manager, I have enjoyed learning how AI can free me from administrative tasks, repetitive work, and help me produce higher quality output. But let's be honest, there were moments where I questioned the whole thing driven by one fear: that I was essentially training my own replacement.

VLMs vs. object detection: A practical use case

Vision language models can describe images in remarkable detail, but when you ask them to tell you exactly where something is in a photo, the results fall apart. This post explores why VLMs fail at spatial localization, what the research says about it, and what we learned firsthand while building an on-device computer vision feature. We compare VLM outputs against dedicated object detection models like YOLO, show the results with real images, and explain why composite pipelines that combine both tools are the most practical path forward.

Designing on-demand ML training infrastructure on AWS: A practical architecture guide

Machine learning projects do not depend only on good models. They also depend on the infrastructure behind them. Training workloads are expensive, unpredictable, and often inefficient when run on static environments. As organizations scale their AI efforts, the ability to provision training infrastructure on demand becomes critical for both cost control and speed of experimentation

Want to see our projects?

Built by Ignit.

EXPLORE WORK