Posted by david927 22 hours ago
Ask HN: What Are You Working On? (March 2026)
It is a pretty fun project
Sounds like a fun project -- like a more interactive version of Football Manager.
One thing I am exploring is that, with LLM (and enough $ to pay the tokens) you can filter out offenses and hate speech in player communications.
All typed message is effectively translated by the LLM as “coach speech” and what is put in the UI is the LLM output. Sure, it is not exactly the player words and can get expensive, but I think it can be a good solution
The problem was the ML dependencies. The backend uses BGE-small-en-v1.5 for embeddings and FAISS for vector search. Both are C++/Python. Using them from Go means CGO, which means a C toolchain in your build, platform-specific binaries, and the end of go get && go build.
So I wrote both from scratch in pure Go.
goformer (https://www.mikeayles.com/blog/goformer/) loads HuggingFace safetensors directly and runs BERT inference. No ONNX export step, no Python in the build pipeline. It produces embeddings that match the Python reference to cosine similarity > 0.9999. It's 10-50x slower than ONNX Runtime, but for my workload (embed one short query at search time, batch ingest at deploy time) 154ms per embedding is noise.
goformersearch (https://www.mikeayles.com/blog/goformersearch/) is the vector index. Brute-force and HNSW, same interface, swap with one line. I couldn't justify pulling in FAISS for the index sizes I'm dealing with (10k-50k vectors), and the pure Go HNSW searches in under 0.5ms at 50k vectors. Had to settle for HNSW over FAISS's IVF-PQ, but at this scale the recall tradeoff is fine.
The interesting bit was finding the crossover point where HNSW beats brute-force. At 384 dimensions it's around 2,400 vectors. Below that, just scan everything, the graph overhead isn't worth it. I wrote it up with benchmarks against FAISS for reference.
Together they're a zero-dependency semantic search stack. go get both libraries, download a model from HuggingFace, and you have embedding generation + vector search in a single static binary. No Python, no Docker, no CGO.
Is it better than ONNX/FAISS? Heck no. I just did it because I wanted to try out Go.
goformer: https://github.com/MichaelAyles/goformer
goformersearch: https://github.com/MichaelAyles/goformersearch
It's an auction website for schools, charities etc without the exploitative transaction fees.
My wife and I are pretty heavily involved in our son's school PTA (parent teacher association) and have helped run school fundraising events for a few years, so we feel sort of like domain experts in this area :)
https://greenmtnboy.github.io/sf_tree_reporting/#/
For all the places it's bad at, AI has been fantastic for making targeted data experiences a lot more accessible to build (see MotherDuck and dives, etc), as long as you can keep the actual data access grounded. Years of tableau/looker have atrophied my creativity a bit, trying to get back to having more fun.
This is an app that’s been bouncing around in my head for over a decade but finally got it working well enough for my own purposes about a year and a half ago.
I was thinking of a google maps kind of "here you are, here's your walking path of interesting trees" potentially, or something else that could tie the overview to the street experience - on the backlog!
Otherwise the map tiles are coming from OpenFreeMap [1] which are indeed based on OSM.
Next steps I'm interested in are including economic + ecological benefits of the trees, highlighting potential pests / invasive species, maybe some other basic info about the species sourced from Wikipedia.
I like how you've got different icons for different types of trees; I've been thinking about how to encode DBH data as well but haven't settled on anything yet.
The first thing I cleaned up was TCL-Edit <https://gitlab.com/siddfinch/tcl-editor>, a small Tcl/Tk text editor I wrote a long time ago. After seeing the Rust clone of Microsoft EDIT, I realized the obvious next step was to build a Tcl/Tk clone of the Rust clone of Microsoft Edit. Recursion shouldn't be limited to code.
I also built a tiny URL system in Perl <https://gitlab.com/siddfinch/perl-tiny-url>, meant to run locally. The idea is simple: short URLs for internal/VPN resources per client. I usually spin up a small daemon (or container) per client and get a dashboard of links I use frequently or recently.
Security is intentionally minimal since it's local, which conveniently lets me ignore authentication and other responsible behavior.
Goal for the year: Continue to open stuff42.tar.gz, pick something, clean it up just enough, and release it, and not have it by the end of the year.
Might even choose a language that might even be described as "modern."
Free Math Sheets is a tool to generate math worksheet PDFs (and the answer keys if required). Currently it supports K-5 but I want to expand it to higher levels of math (Calculus, Physics, you name it!). You select a bunch of different options and then generate it. All in the front-end. No back-end or login in required. https://www.freemathsheets.com
If you are interested in helping out or forking it, here is the github repo github.com/sophikos/free-math-sheets
The paid project is Numerikos. I am going for something in between Khan Academy and Math Academy. I like the playfulness and answer input methods from Khan Academy (but it is linear, doesn't have a good way to go back and practice, etc.). I like Math Academy's algorithm (but it has multiple choice answers, yuck! and is easy to get stuck and doesn't have a good way to explore on your own). Currently Numerikos supports 4th and 5th grade math lessons and practice. The algorithm is based on mastery learning like Numerikos, but you can also see a list of all the skills and practice whatever you want. I am also working on a dashboard system where you can build your own daily/weekly practices for the skills you care about. Next up is 6th grade math and placement tests.
The idea is that _any_ user-facing tool, whether an app, worksheet generator or whatever, will need something like this for content, so I'm making this available for free and hoping for others to build on top of it.
I'm sticking to university-level stuff because I feel that school-level, especially math, is over-saturated already.
Technically, it is currently built as a React app, but that is mostly me sticking to tools that get out of my way. Generating PDFs or Anki files should be relatively straightforward.