Bret Bernhoft

I am a software engineer specializing in low-code development, with a strong focus on building secure, scalable applications using Quickbase and cloud-integrated automation platforms. My background spans from WordPress deployments to enterprise-grade business systems, giving me a unique perspective on how people interact with technology at scale. Since 2021, I have led the development of production-ready Quickbase applications, engineered API integrations across services like AWS, Trello and Zapier and delivered automation workflows to cut operational overhead. My work consistently emphasizes collaboration, user-centric interfaces and real-time reporting to empower all users. I focus on blending application development, business process automation and platform interoperability. My current toolbox includes Quickbase, AI, Python, TypeScript, React, SQL and platforms such as Zapier. I also maintain a private homelab where I prototype integrations with self-hosted LLMs and AI-driven tooling. My ongoing goal is to design systems that simplify complexity, reduce manual effort and expand access to reliable, high-performing technology.

[bretbernhoft]

Earlier this week I shared results from submitting 1,800+ remote technology roles to Anthropic in search of top skills employers hire for in 2026. Which cost $35 to process.

This afternoon I submitted the same 1,066 text documents to a self-hosted AI RAG system on my network. While asking for similar information.

According to Gemma 4 E4B, an LLM from google DeepMind, the most in-demand skill areas employers are looking for in remote workers are technical, soft and specialized. The same LLM mentioned employers value candidates with foundational knowledge (Python, AWS, DevOps) and deep hands-on experience (personal projects, homelab).

While chatting with my self-hosted RAG system this afternoon, I was surprised to notice the following:

1 - Even small LLMs require 100% of a GPU's processing power when answering questions 2 - Indexing 1,000+ documents took 11 minutes, which was longer than originally expected 3 - Locally hosted AI is capable of replacing most cloud-based subscription LLM services

Working with artificial intelligence running on hardware I own has been a game changer for me. Not only is AI inherently transformative, but to be responsible for the whole stack enables true force multiplication.

[bretbernhoft]

After scraping hundreds of websites in seconds, I submitted 1,854 remote tech job postings to Anthropic for analysis of the top skills mentioned therein.

According to this research the top five skills being hired for among currently open remote technology roles are:

1 - Python

2 - AWS

3 - SQL

4 - React

5 - TypeScript

It is interesting to note Python is by far the most in-demand skill for remote tech work in 2026. It could be said the Python programming language holds the software engineering ecosystem together.

[bretbernhoft]

Storytelling with data is a powerful skill.

For example, Calagator is an events calendar for the Portland, Oregon tech community. And has been at the heart of it for 20 years.

What you see in this GIF are the 21,913 events posted to the Calagator website since launching, played back over time. I find these visualizations rather succinct and compelling.

[bretbernhoft]

In the video attached to this post 4 scenes demonstrate a wide range of possibilities. From self-transforming fractals to endless oil paintings, there is a lot going.

[bretbernhoft]

In the video attached to this post you can observe a typical workflow for uploading and indexing documents to my homelab RAG system. Followed by asking a relevant question about the future of software engineering in the context of AI.

[bretbernhoft]

After exporting 593 days worth of traffic logs from my OPNsense homelab firewall, I developed a Python script to process the CSV data into JSON. Which I then visualized using D3.js and JavaScript.

[bretbernhoft]

I am building a colorful little side-scroller game titled "Neon Run". It is built using TypeScript, Vite, JavaScript and zero assets. The game consists entirely/only of code computed at runtime and all player scores are stored in local storage.

[bretbernhoft]

The development of my Ollama-based AI chat interface is coming along nicely.

[bretbernhoft]

I recently developed a computer vision system designed to authenticate spices, starting with turmeric. After training and validating the model on a custom dataset of 203 images, it has already achieved an accuracy rate of over 90%.

[bretbernhoft]

This is what I am dealing with these days! My cat loves to live on top of my homelab cables and power supplies.

I am grateful for both my cat and computers. So this post is more about showing how both coexist.

[bretbernhoft]

This short video was also assembled in less than 10 seconds using the Python programming language. Which means it takes longer to watch the media than it did to create in the first place.

[bretbernhoft]

This fun little video was assembled in less than 10 seconds using only the Python programming language. There was no human intervention in the creative process whatsoever.

[bretbernhoft]

Computer gaming is not collapsing, it is entering another winter season.

On the other side is something far greater than what we know today.

[bretbernhoft]

This is what I deal with every day.

[bretbernhoft]

A short, quickened video showcasing how the Gemma 4 family of LLMs from Google DeepMind can assist with learning how to code for the internet.

[bretbernhoft]

Recent (1) changes to and (2) experiments with the AV-Sync software have produced noteworthy results. Ultimately the software is increasingly capable of matching video clips to audio feature values set between pairs of onsets. If I can crack that lock, this whole world will open.

The entire purpose of this development project is to automate flow state attainment through neural entrainment with specialized media. Such as the video above is an early example of.

When version 1.0 is released, the value this endeavor delivers will be quite clear. Until then, what I am sharing in this thread might seem odd or counterintuitive. That's okay.

Stay tuned-in and you will most likely agree with me, the effort is worthwhile.

[bretbernhoft]

A sped-up conversation held between Gemma 4 (an LLM AI) and myself earlier tonight on two computers in my private homelab.