Your AI Agent Passed the Demo. Now Try to Break It.A tool-using agent can produce the right final answer for the wrong reasons—and once it can change external state, that difference matters.
The ag...
Read moreYour AI Agent Passed the Demo. Now Try to Break It.
Why LLM Evaluation Gets Weird the Moment You Add ToolsA model can produce the right answer and still take the wrong action. Once LLMs can call tools, evaluation stops being only about outputs and s...
Read moreWhy LLM Evaluation Gets Weird the Moment You Add Tools
From Monolith to MicroservicesWhen I started building my final project for software architecture class — a small e-commerce platform — I wrote everything in a single Node.js app.User authenticatio...
Read moreFrom Monolith to Microservices: A Student's Journey with Node.js
Trying to Understand MoE: How LLMs Get Both Bigger and SmarterEvery few months, a new model drops that makes the field collectively gasp.“Trillions of parameters.”It sounds absurd — not just large,...
Read moreTrying to Understand MoE: How LLMs Get Both Bigger and Smarter
Leveling Up in React: My Journey with Advanced HooksFor the first few months of learning React, my world was small and comfortable — defined almost entirely by useState and useEffect. They were the...
Read moreLeveling Up in React: My Journey with Advanced Hooks
RAG Works Great—Until Your Documents DisagreeRetrieving the right documents is only half the problem. The harder question is what the model does when those documents describe different versions of ...
Read moreRAG Works Great—Until Your Documents Disagree
My TypeScript “Aha!” Moments: Best Practices I Learned on a Real ProjectWhen I started building a semester-long project in TypeScript, I thought the type system would automatically make our code bu...
Read moreMy TypeScript 'Aha!' Moments: Best Practices I Learned on a Real Project
My App Was Working, But It Was SlowWhen I first deployed my portfolio project, it looked fine on my machine. The layout worked, the transitions felt smooth, and I was proud of the code.Then came th...
Read moreMy App Was Working, But It Was Slow: A Student's Guide to Web Performance
What Actually Breaks When You Quantize an LLM?Quantization can cut model memory dramatically without obviously hurting benchmark accuracy. The interesting question is what changes before the benchm...
Read moreWhat Actually Breaks When You Quantize an LLM?
A Grad Student’s Guide to Fine-Tuning LLMs: From Brute Force to FinesseWhen I started my first serious research project, the goal sounded straightforward enough: adapt a large, general-purpose lang...
Read moreA Grad Student's Guide to Fine-Tuning LLMs: From Brute Force to Finesse
