Posted by david927 3/30/2025
Ask HN: What are you working on? (March 2025)
I wanted a better way to keep track of applications I sent out, A spreadsheet just seemed like a poor way of tracking data. So overtime I built a desktop application to track my job search activity for me. Most alternatives are web-based, but I didn’t love the idea of broadcasting my job search to third parties. This is a native desktop app (Windows/macOS) that keeps everything local.
Still working on code signing (so no scary "unknown publisher" warnings), but otherwise, v1 is ready.
Would love feedback—especially from others who’ve struggled with job-search tracking!
That and I am able to customize the flow of applications, and the number of swim lanes would be a bit messy. I have my app setup to track interviews, phone screens, take home assignments, and many more.
I found that a smart sorting algorithm is best for displaying the applications.
But its still early days of the app, possibly someday.
https://github.com/Belphemur/night-routine
I wanted to build everything from the ground up in Go and have it fully integrated with Google Calendar where we have our family calendar.
It setup full day event with the name of the parent in charge of the night routine. To override a routine, any of us can just rename the event with the the other parent name and the software recalculate the following routines.
I also wanted to give a try to Roo Code in VS Code it only took me 2 days (evenings) to code the whole thing with a proper sqlite db.
Also, FWIW, I think I'm the one that is in the deficit of fair although dad's usually get a bad rep in this regard. I end up doing a lot more night time reps because she does frequent girls nights and has multiple friend groups she's trying to stay engaged with and is in a theater group and mahjong group, etc. However, I balance it with my occasional "take an entire day" to myself. Stuff like this is hard to track and why I think it's important to shoot for what 'feels fair' and make sure you talk about it occasionally so nobody suddenly has repressed feelings of inequality.
We wanted both a system that keeps track of it for us. We want to be sure we can both have activities while still not leaving the other parent on the side and rely on feeling.
In our case we also have recurring events like sport in the evening that happens every week at the same time so this help plan around it and not become unbalanced. We already put everything in the calendar :)
The big debate in my head right now is whether a next byte prediction architecture is better or worse than full sequence prediction.
The benefit of next byte prediction is that we only expect 1 byte of information to be produced per execution of the UTM program. The implication here is that the program probably doesn't need a whole lot of interpreter cycles to figure out this single byte each time (given reasonable context size). However, the downside is that you only get 256 levels of signal to work with at tournament selection time. There isn't much gradient when comparing candidates on a specific task.
The full sequence prediction architecture is expected to produce the entire output (i.e., context window size) for each UTM program invocation. This implies that we may need a much larger # of interpreter cycles to play with each time. However, we get a far richer gradient to work with at fitness compare time (100-1000 bytes).
Other options could involve bringing BPE into my life, but I really want to try to think different for now. If I take the bitter lesson as strongly as possible, tokenization and related techniques (next token prediction) could be framed as clever tricks that a different computational model could avoid.
One consistent annoyance in my professional work has been dealing with PDFs – specifically, extracting information into editable formats without losing structure. Copy-pasting often creates a mess.
So, my first project tackling this is an online PDF to Markdown converter: https://pdftomarkdown.pro/
I've focused heavily on trying to maintain good formatting for headings, text flow, formulas, and especially table structure (getting rows/columns right in Markdown). It also has an online editor for quick modifications after conversion.
A key aspect for me was privacy: the application explicitly does not save the content of uploaded PDFs or the generated Markdown files. It only stores minimal metadata (email, filename, page count) for registered users' plan limits.
It's very much a "scratching my own itch" project born out of that PDF frustration. Early days, but hoping it proves useful for others too.
I couldn't find a way to contact you, so if you feel like it, drop me an email (email on my website in profile).
The need for batch processing to pull out targeted data points from PDFs (rather than converting the whole document) is a valuable insight.
While the current tool focuses on full conversion to Markdown, enhancing https://pdftomarkdown.pro/ to handle specific data extraction tasks like yours is definitely something I'll consider carefully for the future roadmap. Thanks for highlighting it!
The goal is to allow anyone to know how much any car is worth at a given age/mileage, and eventually help people make better purchasing decisions.
Be warned: It's still a very buggy prototype at this stage and the data confidence for all but the most popular models is low!
With the recent flurry of historic events unfolding, I want to see it from different perspectives (e.g. U.S., Europe, Russia, China, pro-Palestine vs Pro-Israel), therefore I included channels from all these areas, even channels that may be considered propaganda. So keep a critical eye when watching them.
And it's a way for me to try out Vidstack and SvelteKit. Feels like the routing can be improved though.
Which I think is a valid answer. I have a job, family and some health issues.
The main thing I am looking at is blogging. Just posts on a problem I solved that week at work kind of thing. Seems like a low time cost way to promote myself. I might dip my toe in the cooking fat of linkedin engagement using the posts.
As much as that is yuk I feel it may be beneficial. Just need one linked in lurker to be impressed and hire me in a few years time!
And by not posting about AI or working 200 hour weeks I might stand out :)
Will continue to refine and possibly do more - love iterating and polishing as a way to learn.