Articles
Long-form explainers, short technical notes, and paper commentary across Towards Data Science and X Articles.
Article Links
How to design experiments that evaluate agentic harnesses, not just isolated model calls.
A deep dive into exactly how text-only language models are fine-tuned to see images.
A practical method for building agentic harnesses: scoped workflows, tool boundaries, evaluation loops, and failure handling.
Step-by-step guide to building autonomous memory retrieval systems.
A compact explanation of Recursive Language Models and why recursive structure can help with long-context reasoning.
An introduction to diffusion language models: how they generate text differently, where they fit, and when they are worth considering.
A practical explanation of speculative decoding for LLM inference: draft models, verification, and why it can speed up generation.
Simplifying the concepts required to master reinforcement learning.
A practical introduction to deep learning internals through PyTorch.
A hands-on walkthrough of context engineering, one module at a time.
A visual tour and from-scratch guide to train GRPO reasoning models in PyTorch.
Residual vector quantizers, conversational speech AI, and talkative transformers.
A visual tour of what it takes to build strong retrieval-augmented LLM pipelines.
Simplifying the neural networks behind generative video diffusion.
How foundation, promptable, interactive, and video segmentation fit together.
A paper review of Self-Distillation with Policy Optimization, covering how models can improve by generating and learning from their own preference signal.
A review of Minimax M2.5 post-training, with emphasis on asynchronous RL, scaling training throughput, and reasoning-model optimization.
A paper-review breakdown of GLM-5 training: pretraining, post-training, reasoning behavior, and the engineering choices behind the model family.
A paper review on training smaller deep-research agents that can search, synthesize, cite, and iteratively improve research workflows.
A review of HyperAgents, a Darwin-Godel-inspired framework for agents that evolve skills, tools, and strategies through self-improvement loops.
A paper review of self-improving AI systems that combine auto-research, code evolution, and harness-level feedback to improve agent behavior.
A review of Google's SkillOS work and the role of skill curation in making self-evolving agents more reliable and reusable.