How close does Jev get to task-specific models?
An empirical comparison of general-purpose zero-shot inference with task-specific supervised models.
Archive
Projects, experiments, writing, and notes from over the years. Dates mark when each artifact was first published or made public.
2026
Jev Benchmarks
Reproducible comparisons of Jev against supervised and zero-shot baselines across five NLP benchmarks.
Sokoban Solver
A push-based breadth-first search solver and visualizer for Sokoban puzzles.
Making LLMs Simpler: From SFT to GRPO
Extending a supervised fine-tuning experiment with reinforcement learning for simpler technical answers.
SQL Agent
A minimal LangGraph ReAct agent evaluated on 250 natural-language SQL questions.
Self-Hosted Hermes
Lessons from using a personal AI agent for learning, household workflows, and small everyday problems.
LLMs in 10 Words
Complex language-model concepts explained in ten words or fewer.
GPU Chip Cheat Sheet
A quick guide to decoding GPU names and comparing the specifications that matter for a workload.
Get LLMs to Say More with Less
Fine-tuning a small language model to produce answers that are both correct and simple.
Simple LLM
An experiment in post-training Qwen3.5-4B to write clear, correct technical answers in simple English.
LLM Koans
Small, focused PyTorch exercises for learning how large language models work.
Personal AI Wiki
A growing, linked knowledge base built from AI newsletters and personal research.
2020
End-to-end Machine Learning with TFX on TensorFlow 2.x
A walkthrough of building an end-to-end production machine-learning pipeline with TensorFlow Extended.
Data Science Reading List
The practical data-science reading list I wish I had from day one.
2019
Andre's User Manual
How I think, collaborate, communicate, and what I value when working with a team.
2017
What Deep Learning Taught Me About Competition
Using agent-based models and GANs to examine how competition and cooperation can coexist.
What Hackathons Can Teach Us About Running a Startup
What rapid prototyping and continuous feedback at hackathons reveal about building products.
2016
Dev Environment
A reproducible setup for carrying a personal development environment across Linux and macOS.
DMV Stalker
A bot that monitors scarce driving-test appointments and sends timely Slack notifications.