I built a multi-task NLP system to do summarization, emotion detection, and topic classification from one shared encoder as an undergraduate research project, implementing the transformer from scratch and training it on my personal 4070. In April 2026 I was preparing it for submission when I audited my own draft and found that several baseline and ablation numbers were not good enough. I withdrew the paper and did a writeup on Substack. The verified results are there. What I took from it is that the evaluation was the hard part, not the model. Getting honest numbers out of a 28-class multi-label problem with a long tail where threshold choice, split reuse, and metric selection can each move the headline by ten points was harder than getting the training work for the model, and it's where the original draft went wrong. I'm now rebuilding it around a question I think is more useful: given a fixed post-training compute budget, is it better to jointly train one multi-task model, train separate specialists, or merge those specialists? Training is paused while I rebuild the data sources. LexiMind also has a live book-discovery site now, with search and "more like this" recommendations over 120 sourced Open Library works.

Two products I'm building under one name. Study takes an ML concept through six stages, from the big picture and intuition through the maths and implementation to a recall check, so you understand it well enough to retrieve it later. Studio turns ML ideas into interactive figures: start from a preset, edit the data and the explanation, and export a PNG, SVG or HTML file for an article, slide or doc.