• My kid played a dinosaur game inside my agent's reply. The hard part was keeping it alive.

  Oct 2, 2026

  My personal agent can now answer with a small web page, drawn in a sandboxed frame inside the chat bubble. Getting a game to show up was easy. Keeping it safe, on-theme, and running when the next message arrived was the actual work.

• Twenty-one raises, then never up again. The budget got so tight it made workers delete the wrong things.

  Sep 23, 2026

  A one-way ratchet on code size stopped my agent-written codebase growing. Clamping it exactly onto the tree turned out to be the mistake, and the fix is three hundred lines of deliberate slack.

• My worker agent finished the job. A 1 MiB buffer ate the answer.

  Sep 21, 2026

  An orchestrator that talks to a human and dispatches long-running workers has an IO problem that is harder than its scheduling problem. Every design mistake I can point at turned out to be an IO mistake.

• My agents proved eleven items done in sixty seconds. None of them were.

  Sep 21, 2026

  Two things make agents ship quickly — a work board a human can actually verify, and a gate cheap enough that everybody runs it. Both turned out to be the same problem, and it is a design problem rather than a discipline problem.

• 1,169 times an agent rewrote itself. Here is what it changed.

  Sep 17, 2026

  A taxonomy of every self-edit an autonomous agent made over seven weeks, with a pie chart, and the one category that ever made the codebase smaller.

• The reply was correct. Nobody could tell.

  Sep 17, 2026

  Two replies to my agent carried the same facts. One got a thumbs up, one got a thumbs down, and the difference wasn't accuracy — it was whether the verdict survived contact with the reader.

• Your agent spends ten seconds a reply drawing table borders

  Sep 17, 2026

  I measured what formatting costs an LLM agent. A quarter of every long reply is pipes and dashes, generated one token at a time while someone waits.

• I gave an agent the ability to improve itself. It got very good at improving itself.

  Sep 17, 2026

  Four mistakes from building a self-evolving agent, and the one measurement that would have caught all of them.

• Engineers will stop writing code. That is not the same as being replaced.

  Sep 17, 2026

  Hand-typing code is becoming the slow part of software. I think the job moves to design, review and the gates you set for agents — and here is the best evidence against that.

• I solved the same problem twice, in opposite directions

  Sep 17, 2026

  Version one put a language model in the reply path to format replies. Version two deleted it and wrote a parser. Here is what each one actually cost.

• A non-blocking agent orchestrator

  Aug 31, 2026

  Almost every multi-agent framework blocks the orchestrator while its subagents run. Here is what changes when you remove the ability to block instead of managing it.

• Pax Layer Basics

  Apr 28, 2023

  This lab describes the basics for authoring a new Pax layer.

• Building a classifier using JAX

  Apr 2, 2023

  Creates a synthetic dataset for a linear classification task and trains a logistic regression model using JAX, FLAX and OPTAX.

• Using TPU runtimes in colab for JAXs

  Mar 29, 2023

  Explains how functools partial works with simple examples.

• functools partial

  Oct 27, 2022

  Explains how functools partial works with simple examples.

• Building a simple java app with external dependencies

  Apr 13, 2022

  This blog describes, how we can build a simple java application using couple of external dependencies. This can help in decopuling certain components from a big corporate application and test it.

• Installing tensorflow on M1 Macs

  Apr 13, 2022

  Step by step instructions for installing tensorflow on M1 Macbooks with Apple Silicon.

• File transfer from blob storage using azure cli

  Jun 2, 2021

  Step by step instructions on how to move files and directories to and from azure blob storage.

• Runing ML Training code on a VM

  Dec 9, 2020

  How to move a local machine learning traning code to a virtual machine and develop there.

• Train a Covid19 Tweet sentiment classifier using Bert

  Dec 8, 2020

  End to end training code for a bert model using pure pytorch and a single GPU.

• Automatically activate conda evironment in Powershell for VSCode

  Sep 17, 2020

  How to set up VSCode to automatically activate anaconda environment for powershell.

• Host python code documentation using azure app service CI CD pipeline

  Sep 17, 2020

  How to upload python static html documentation created using mkdocs to an azure app service and enable CI CD pipeline using azure devops.

• Visualizing Bert Embeddings

  Aug 27, 2020

  Visualize bert word Embeddings, position embeddings and contextual embeddings using TensorBoard

• Using PyTorch 1.6 native AMP

  Aug 4, 2020

  Showcasing how to use native amp's autocast() and Gradscaler through simple example model.

• Zero shot NER using RoBERTA

  Jun 25, 2020

  Using Roberta last layer embedding and cosine similarity, NER can be performed in a zero shot manner. The model performance is very good without any training. This notebooks finds similar entities given an example entity.

• Type faster using RoBERTA

  Jun 10, 2020

  Typing fast on phones are prone to error. Using masked language models, these errors can be detected and rectified

• Using Tensorboard efficiently in AzureML

  Apr 28, 2020

  Using tensorboard client effeciently in azureml

• Using Tensorboard in Pytorch

  Mar 24, 2020

  Using tensorboard in pytorch. This example uses windoes for the system commands. Linux and Mac will need slight modification in the powershell commands

• Using Tensorboard in Tensorflow-Keras (windows version)

  Mar 23, 2020

• Powershell bashrc equivalent

  Mar 3, 2020

• Resize image using Python

  Feb 17, 2020

• Issue with Gradient accumulation while using apex (Fix included)

  Feb 6, 2020

  Apex with gradient accumulation results in inaccurate gradients if reduction is disabled for steps where optimizer is not stepping.

• NVIDIA mixed precission training

  Jan 20, 2020

  A clear understanding of how NVIDIA mixed precission training works. This blog also lists out official documentations necessary to understand the concepts

• Undo a git rebase

  Jan 20, 2020

• Challenges of using HDInsight for pyspark

  Jan 12, 2020

  Challenges and approaces to submit job in VS code with links, results and opinions

• Insertion transformer summary

  Jan 6, 2020

  Insertion transformer paper breif summary

• Spark Quickstart on Windows 10 Machine

  Jan 3, 2020

  How to install and run spark in local windows 10 machines using two installation methods - binaries and pip.

• PyTorch distributed communication - Multi node

  Oct 15, 2019

• Bert Attention Visualization

  Sep 26, 2019

• How to create a new docker image

  Sep 23, 2019

• LAMB paper summary

  Sep 21, 2019

• Bert Memory Consumption

  Sep 20, 2019

• Introduction to Transformers

  Sep 1, 2019

• Contingency table and Chi-squared distribution

  Aug 26, 2019

• Azure Machine Learning Tutorial

  Jul 8, 2019

• PyTorch IsRead Predictor on my email

  Jun 27, 2019

• PyTorch BERT

  Jun 26, 2019

• PyTorch RNN

  Jun 20, 2019

• Word analogy using Glove Embeddings

  Jun 13, 2019

• Convolution Explained

  Jun 13, 2019

• Activation and Loss function implementations

  Jun 6, 2019

• How to use Jekyl!

  Dec 1, 2018

• Python Flask web application in azure linux

  Dec 1, 2018

  Start a python flask web applicaiton in azure linux machine and understand the basics behind that.