Why did I make this website and what to expect
There are a few different reasons I built this blog website. I recently completed the Blue Dot Impact Technical AI Safety Course and one of the last classes was partially about academic and research communication. Aside from this AI Safety related work I have done many other projects and as part of the course and as part of my previous job and hobby projects, I have read many research papers. Reading them or doing projects in isolation is usually not enough so I plan to put different parts of the projects in here and sometimes add in my thoughts about some of the papers and assignments I did as part of the course.
In addition to just sharing information, I believe this approach could help me think better and for some projects it could serve as an explainer for myself as to why certain things were done. For example the post after this one will dive deep into the experimental design for my local LLM tool use verification project. Another post will discuss the results in depth, and another will discuss how logging was done and some pitfalls.
I will also add other projects and their details. For example building this website was actually the first time I built something only using LLMs, specifically Claude Code in this case. But I have some background in software security and in AI safety so I was very skeptical of directly installing it on my machine. That post will talk about why I used the web version of Claude Code instead of the CLI or the VSCode extension.
There might be another about using local LLMs or mixing local LLMs with cloud based LLMs for some use cases.
AI policy for this blog
The posts will be written "by hand" by myself because I need to keep full control over the main ideas and I don't want LLMs to change something that I wanted to convey. That said, at this point I am not completely comfortable with my own writing so sometimes I might go back and fourth to refine my prose using Claude or another large LLM. Maybe later switching to a local one instead and put it as a pre-publishing tool within the blog to correct more grammatical type of errors or polishing the writing in general. Basically things that a regular spellcheck or grammar check can't fix.
Why not use Linkedin or other pre-made services
I will be posting on Linkedin as well but it doesn't give as much control over formatting. There are also problems with very larger posts that people may not want to read so Linkedin will get a smaller summary instead.
Technical side for this blog
This blog is made using Claude Code and Django. I have more experience using FastAPI than Django but the reason for using Django here is because of the "batteries included" part. Django already comes with an admin panel, it comes with SQLLite ready to test the database, the model system is all pre-built and the templating enging handles views without much additional work. This is the kind of use case where Django excels. There are of course downsides to it, because it is much more openionated, it is a much "heavier" framework in terms of disk use and memory use. I would not use this for something like an API but for a blog website, Django is great.