This starts from my frustration when opening large CSV files, but later evolved to a log/data analyzer that loads arbitrary format in constant time (O(1)).

https://github.com/Verticalysis/Hitomi

The secret: I engineered an incremental combinatorial parser capable of processing customized format from a steam. Any inputs, including file or the stdout from a command, are first chunked and then fed to the pipeline. The UI is ready when the first small chunk is processed.

Other highlights: 2-mode filter, one with a convenient UI and the other is based on an extensible DSL for complex cases;

Timeline mode scrollbar, a secret weapon for log or time series analysis;

Column widths fit to content automatically;

Native code, no web bloat;

Cross-platform (currently Windows and Linux, MacOS WIP).

If you are tired of all the quirks Excel have when working with CSV files, you'll gonna love it!

Building a GNSS/RTK-based lawnmower.

The existing ones were quite expensive, especially when I started out. A friend had the idea to get a cheap/non-functioning lawnmower second hand, and tear out the circuit board. We're in the process of coding up a new ROS2 based stack that will roam the lawn on GPS with RTK in the charging station. My friend does most of the electronics stuff, and I focus on the software.

I'm at the point where I will start testing a simple bounding box soon and just have it drive around until it "hits the edge" and then randomly pick a new direction.

It's fun so see the software I build "in real life" instead of as a web-site, as is the case for my my daily job.

I am working on a minimalist journal app that is for really quick single line jots. Can be used for idea dumping, project management, quick calculations, unit conversion, task management etc. Have unlimited undo/redo support, assign tags etc. Will be adding scripting support using Python next. The data is stored in text file, in really human friendly manner, but also in way so that the *nix tool users can easy to navigate the file using text processing tools.

Native application, no web UI, built using Rust + iced.rs, minimal dependency. NO AI.

I am putting the best effort to make it performant. Target audience is the users who want's the simplicity of the notepad [non-sloppy one], but still with some bells and whistles to note without worrying about managing the metadata manually.

I think with scripting there will be infinite possibilities to play with linear notes, and I want to make that happen.

Continuous challenges while implementing features are:

  1. It should load instantly
  2. Keeping it extremely simple to use
  3. Keeping the interface minimal
  4. Still have ways to let the user find the features easily.

Will have a demo version ready soon

I've shared this one before but I built a logistics management system to power deliveries for a business I founded years back and I've continued to refine it since:

https://toanoa.com/

Since the initial MVP, it's done close to 100k orders and I've added new functionality like:

- Intelligent order batching & route optimization that can interleave tasks across orders in such a way that they still have the best chance possible of completion within their delivery windows

- Further refined the mobile tracking logic in our driver app to improve the quality/frequency of position updates while continuing to be as efficient as possible on battery

- Numerous backend/DB optimizations such that average response times are in the tens of ms at the current volumes it's handling.

It's not open source but if you have an interesting use case and are curious about it, feel free to reach out.

I am trying to build a simulation that lets a simulated organism come up with its own small language, purely learned from sensory input: https://github.com/JoergStrebel/VirtualZoo/blob/main/compute... I would like to implement the ideas put forward by Stevan Harnad in his symbol grounding problem paper (Harnad, 1990).

I’m building a blackjack card counting tool for people to learn how to count and how to identify games that are winnable. It is designed to take a completely novice to an advanced, winning card counter, using a Duolingo like approach - mastery based learning across sequential modules. Minus the ads and dark patterns.

A better permissions layer for coding agents. The tool works like auto-mode for Claude Code, so you can stay in the flow and only get prompted to allow or deny tool calls when it truly matters, but it is fully deterministic. My benchmarks surfaced that most Bash calls don’t need an LLM to be classified as safe, ambiguous, or dangerous. A deterministic classifier can auto-allow or block 95% of Bash tool calls as safe or dangerous, with only the remaining 5% being truly ambiguous or unknown.

Conclusion is permission reviews with LLMs like Claude’s auto mode or Codex auto review are like using a data center to flip a light switch - overkill.

The main benefit is that your agent’s autonomy can be governed deterministically through policies that can be stored at the user and repo level. The bonus is that you save tokens vs using auto modes.

https://nah.build

I've been cataloging my collection of American vintage ads (https://adretro.com). The collection has expanded considerably in the last year. I'm working on a front end search portal for users to use, to deeply search all aspects of the collection. Built on MySQL and Lucee. Meta data is extracted with OpenAI image, stored in MySQL FTS. I'll probably add vector search after I get it live.

I've been working on RVW, my adaptation of the standard transformer model that is capable of online continual learning without catastrophic forgetting. I finally published the first pre-print of my early experiments: https://doi.org/10.5281/zenodo.20064617

Now I'm working on expanding the work into more parameters and improving performance. I just finished an extremely harsh test of a Nemotron-flavored RVW that consisted of stretches of a random assortment of domains interspersed with long runs of single domains. Across all of it the model didn't forget (and actually improved on some of the more challenging domains). PPL on SmolTalk is still in the ~18 range, which I'd like to get lower, but this is all with only 4B params.

Currently, I'm training a Llama 3.2-flavored RVW with only about 2B params to see how that turns out. Depending on results of that, I may take it to Gemma 4 next.

After quite a few years of coming up with and implementing 'great ideas' but not being able to follow through to making them revenue generating products, I'm on my best bet so far.

I always wanted to build a real-life puzzle game, which is app/mobile assisted. Had yet another eureka moment, and built a usable prototype (backend plus iOS app). Good feedback from a small circle.

For a while I was aware of someone (I knew by sight) who worked in the same sort of subject matter (but a non-tech). I approached her, we had a coffee, I pitched the idea and how she could bring it to life, as I made the tech side. She jumped on board.

We're two and a half weeks in, have gone full speed and are making something great (for our audience). My future co-founder is amazing, great insights, opinions, drive. We're potentially launching in a couple of weeks, a free/MVP version of a puzzle game.

I've been through many iterations of trying to get something off the ground. Tried tech co-founders, and the last years of going solo (very hard after you've done the coding). But this now feels right. A puzzle app/game for every day people to have some fun. And a future co-founder whose life is outside tech, who's bring a sort of fun energy outwith let's make loads of money or isn't the framework/AI cool.

Balance is good. Contact with reality is good too :)