How the Tool repo on my github works

First off, the tool repo on my github was shamelessly stolen from Simon Willison, who is a prolific software engineer and influential tech blogger. I’ve been following his blog for a bit now and he is excellent at showcasing and explaining what he does.

His tools\ and research\ repos are great vehicles for experimenting in this golden age of coding and I saw it as a great way to ease my own transition back into coding in this new world.

To that extent, I’ve created my own tools\ and research\ repos so I could understand how they work. This is an attempt to document it because I forget things!

Solving puzzles

Solving Advent of Code puzzles is fun! and a great way to learn a new language. I started the 2025 puzzles along with everyone else, when they were released on 1-Dec-2025. It was my first foray into doing them and it felt fun to be part of a collective effort, learning and comparing notes. The puzzles are not easy, especially if you are out of practice! I mean, the first puzzle for each day is not so bad, but each day has 2 puzzles, the second of which gets unlocked if you solve the first puzzle; and the second puzzle is way more challenging than the first.

When AoC (Advent of Code) started, I was in the midst of re-skilling myself with NeoVim and keyboard shortcuts, tmux, etc. I wasn’t ready to go into Clojure territory just yet, so I tackled the puzzles in Python which I’m much more fluent in. However, I only lasted 4 days because (just because).

Now a month later, I’m more comfortable with Tmux and Neovim; I have Clojure and Conjure setup as well as LSPs, parinfer, and all that good stuff. Heck, I even subscribed to Claude Code Max! So now I’m a code warrior, I suppose. Thanks to holiday specials, I got Programming Clojure, 4th Edition at a discount so I’ve been reading that as well.

Day 5

I started on the puzzles for Day 5 and whoah. I know the puzzles get progressively harder, but my first solution just didn’t cut the mustard. I used sets and the problem with that was that the ranges used were gigantic! Take for example: 81344892339775-81647378719492 as just one of the sample inputs. This is HUGE. I used ranges to represent these, but then by using sets, I was asking Clojure/Java to ‘realize’ each individual element in that range, (and do this for a bunch of similar ranges!). Each Java Long takes ~24 bytes, so if I had a billion elements, it would require 24 GB of memory!

So my solution ran out of memory. Also, if I were to do this in Python, it would have resulted in the same memory error.

I wasn’t sure how to go about it without sets so I consulted with my various AI buddies and gave me the idea of using ranges instead, simply denoting the start and ends of each range. Then using math to see if each ingredient was within one of the ranges. Much faster!

One technique is to emply laziness, which I thought I was doing, but had no idea what impact using sets had. I didn’t know about laziness in general, so this was a very good exercise for me. In fact, I had just covered the lazy sequence chapter in the book at just about the same time, so it all came together!

Needless, to say this first puzzle threw me for a loop, but I got the answer. That’s when I got to Puzzle #2.

Here, you need to basically merge ranges together. I sketched out a solution on paper and thought it could work, but it was not going to be easy. In fact this is where my tendency to commit too early to a solution when solving problems was leading me astray (as opposed to considering a wider array of approaches initially).

I fed my approach to AI and it was very good at coaching me out of it! It gave me edge cases where my algorithm would turn into a mess. It was do-able but it was going to get quite complicated. Instead it suggested a much more reasonable path forward. This path included sorting the ranges from low to high and then walking through them and merging overlapping ones. Each time I wrote some code, I fed it into AI and it gave me more idiomatic ways to write the code. So that was another great way to learn.

So Day 5 is done. I will proceed with the other days in a similar fashion. Hopefully my Clojure becomes more fluent and idiomatic moving forward. However, the problems themselves seem like they’ll become more challenging!

I also want to go back to Days 1-4 which I did using Python and do the second puzzles for those days as well.

UPDATE

I made it to Day 10 and then went on vacation. And when I got back, other things came to the fore: notably AI fever! I realized I needed to learn/avail myself of these new tools, and that hand-coding was going to become akin to working on a custom car in your garage.

Clojure and thinking functionally is beautiful and elegant and something I want to get back to, but for now, I’m pivoting to exploring AI and its myriad uses…

Set up of Skills in Cursor

This post details the steps to get the skills capability running in Cursor which I have a year-long subscription to and want to maximize usage of before having to pay for Claude or OpenAI.

I had no clue you could create dotfiles to memorialize your setup! And to be able to recreate it on other machines.

This whole exploration into Python async driven by LLM coding has been really interesting. I wanted to understand this better as I couldn’t really grok it. My mental model is still treating an async for loop as just a regular for loop but there are differences!

I was trying out various tutorials to learn AI, specifically Google’s ADK and a course on Gradio at Deeplearning.ai; both courses featured code in notebooks that sourced from APIs and both seemed like they did streaming. But I wanted a deeper understanding of how this mechanism worked. Cuz I’m old and I’ve been out of the coding game for quite some time so have little to no experience with streaming.

So I actually vibe coded a site and put it into production. It wasn’t a terribly sophisticated site. It was a single page application on a Typescript/React stack with no database. Really pretty simple.

But it was a passion project and I needed it up and running in a very quick timeframe. It was a lot of fun because basically everything was new to me. I used Cursor for this development and made more use of GitHub than I ever have previously.

Privacy with Email

Today I was playing with the hugging face API when I was following the gradio course on deep learning.ai ; in that course they use the python request API to make calls out to hugging face’s inference API;

You’ll find this post in your _posts directory. Go ahead and edit it and re-build the site to see your changes. You can rebuild the site in many different ways, but the most common way is to run jekyll serve, which launches a web server and auto-regenerates your site when a file is updated.

Lorem ipsum dolor sit amet, consectetur adipiscing elit, sed do eiusmod tempor incididunt ut labore et dolore magna aliqua. Ut enim ad minim veniam, quis nostrud exercitation ullamco laboris nisi ut aliquip ex ea commodo consequat.

Curabitur pretium tincidunt lacus. Nulla gravida orci a odio. Nullam varius, turpis et commodo pharetra, est eros bibendum elit, nec luctus magna felis sollicitudin mauris. Integer in mauris eu nibh euismod gravida. Duis ac tellus et risus vulputate vehicula. Donec lobortis risus a elit.