Readit NewsThe idea:
1. apply strict schema validation to YAML so AI won't generate wrong/invalid data 2. write prompt to AI help people generate a sample YAMLResume 3. adopt AI to parse existing PDF or images and convert it to YAMLResume format
Maybe play around with having OpenAI convert existing resumes into the yaml format to make it easier to try out.
I don't have much experience with modern finetuning, but isn't it highly technical. How many layers do you want to change? What is the learning rate? Does that need to be visible to the user? How many examples are needed in practice
I've been trying it on an M2 64GB via both Ollama and MLX. It's very, very good, and it only uses ~22Gb (via Ollama) or ~15GB (MLX) leaving plenty of memory for running other apps.
Some notes here: https://simonwillison.net/2025/Apr/19/gemma-3-qat-models/
Last night I had it write me a complete plugin for my LLM tool like this:
llm install llm-mlx
llm mlx download-model mlx-community/gemma-3-27b-it-qat-4bit
llm -m mlx-community/gemma-3-27b-it-qat-4bit \
-f https://raw.githubusercontent.com/simonw/llm-hacker-news/refs/heads/main/llm_hacker_news.py \
-f https://raw.githubusercontent.com/simonw/tools/refs/heads/main/github-issue-to-markdown.html \
-s 'Write a new fragments plugin in Python that registers
issue:org/repo/123 which fetches that issue
number from the specified github repo and uses the same
markdown logic as the HTML page to turn that into a
fragment'
From a dev perspective this area has a ton of super interesting algorithmic / math / data structure applications, and computational geometry has always been special to me. It's a lot of fun to work on.
If anyone here is interested in this as a user, I'd love for any feedback or comments, here or you can email me directly: tyler@vexlio.com.
Some pages the HN crowd might be interested in:
* https://vexlio.com/blog/making-diagrams-with-syntax-highligh... * https://vexlio.com/solutions/state-diagram-maker/ * https://vexlio.com/blog/speed-up-your-overleaf-workflow-fast...