Insights on AI, machine learning, and technology trends
Filter by tags:
MCP's 2026-07-28 spec goes stateless — killing the session handshake, sticky pods and long-lived SSE. What that unlocks for agentic e-commerce checkout.
A pragmatic comparison of my hand-rolled Vercel AI SDK multi-agent harness and iii's worker-based harness substrate: what each is good at, where each hurts, and what I learned testing the migration.
I'm a Python person. I built a Rust LLM gateway with Claude Code. Here's why I added another box to the LLM-gateway shelf, what makes Aura agentic-native instead of just OpenAI-compatible, how to use it from your existing framework, and what I learned about vibe engineering production infrastructure as a non-Rust developer.
Everyone is building agents. Almost nobody is building what makes agents defensible. We explore why the agentic harness and data flywheel — not the LLM — are where durable competitive advantage lives in 2026 and beyond.
A deep technical dive into how the open-sourced X/Twitter recommendation algorithm works, and how you can apply the same architectural patterns to build powerful e-commerce recommendation systems.
The industry says AI is forcing engineers to choose between becoming Super ICs or Product Engineers. But there's a third path emerging—the Player-Coach at Scale—that refuses to accept this false dichotomy.
Nowadays, anyone can be a software engineer and build a frontend or backend or entire full-stack application in the age of GenAI and LLMs. However, few can build software that sells, are loved by 1000s and displays good design and taste. In this post we explore why taste might be more important than ever 2025 and beyond.
It is an understatement to say that AI, particularly the Data Science profession, is changing drastically with the introduction of GenAI and Large Language Models. In today's GenAI-driven world, being a good data scientist is more challenging than ever.
Using DSPy for extracting metadata data using NER - Learn how to use DSPy framework for Named Entity Recognition tasks instead of traditional prompt engineering approaches.
Prompt Engineering Tips & Tricks for Named Entity Recognition (NER). Learn how to effectively extract entities from text using Large Language Models with practical techniques and examples.
MCP's 2026-07-28 spec goes stateless — killing the session handshake, sticky pods and long-lived SSE. What that unlocks for agentic e-commerce checkout.
A pragmatic comparison of my hand-rolled Vercel AI SDK multi-agent harness and iii's worker-based harness substrate: what each is good at, where each hurts, and what I learned testing the migration.
I'm a Python person. I built a Rust LLM gateway with Claude Code. Here's why I added another box to the LLM-gateway shelf, what makes Aura agentic-native instead of just OpenAI-compatible, how to use it from your existing framework, and what I learned about vibe engineering production infrastructure as a non-Rust developer.
Everyone is building agents. Almost nobody is building what makes agents defensible. We explore why the agentic harness and data flywheel — not the LLM — are where durable competitive advantage lives in 2026 and beyond.
A deep technical dive into how the open-sourced X/Twitter recommendation algorithm works, and how you can apply the same architectural patterns to build powerful e-commerce recommendation systems.