Posts

I tried building an LLM coding assistant using GPT4. This is what I learned about the current limits of this tech. The main challenge is handling projects that exceed the LLM’s context size, a problem familiar to other LLM applications like document Q&A. This is especially challenging when each query needs to use the most recent data, and the data is updated frequently. Fine-tuning the model for every request isn’t feasible, so an alternative data access solution is required....

What’s your go-to engineering interview question? This one changed my hiring game.

Introducing my latest FOSS project- Machate. Machate is a thin Electron wrapper around OpenAI’s ChatGPT website, transforming it into a desktop app. Residing in your menu bar, it can be accessed quickly via a keyboard shortcut. It takes inspiration from the seemingly abandoned chatgpt-mac and adds much-needed improvements like configurable hotkey and always-on-top. Since this is just a wrapper around the real website, you don’t need any OpenAI API key to use it....

Introducing my latest FOSS project- CodeZen. I regularly utilize LLMs for querying my codebases and have been seeking an efficient solution. The process with ChatGPT can be quite tedious, requiring the manual copying and pasting of file contents. Meanwhile, Github Copilot Chat (beta) falls short when the queries need information from multiple files. This is why I created a tiny CLI tool named CodeZen. It lets you ask a question and sends it to GPT with the entire codebase as context....

The challenge of selling developer productivity tools often lies in accurately measuring their impact. Many teams don’t measure developer KPIs at all, and those who do find that these metrics are highly variable. This is especially true for smaller teams with lower task flow, resulting in smaller sample sizes. Some reasons for varying development times might include: Incomplete or unclear specifications – bad product specifications require the engineer to complete them himself....

In the past few days, there’s been chatter about a perceived decline in the ChatGPT experience with GPT4, particularly when used with plugins. One theory is that these plugins might pollute the initial prompt, which is typically highly optimized, thus leading to subpar outcomes. If this theory is accurate, we should be careful about the plugins we integrate with our AI agent. More connected plugins might lead to worse results. Different plugin combinations could yield varying results....

To unit test or not to unit test? This post offers an efficient compromise.