Log - field notes

Blog

Notes on models, signals, and the parts of research that do not make it into papers. Mostly things I wish someone had told me earlier.

All entries

N = 13
2026.05.15 tools 5 min

Build the Thing You Keep Wishing Existed

The best tools come from being annoyed often enough. On scratching your own itch, and why researchers should write more software.

open →
2026.05.08 research 5 min

Future You Is a Stranger. Be Nice to Them.

Reproducibility is not about virtue. It is about the fact that in six months you will not remember anything, and a reviewer will ask.

open →
2026.05.01 personal 6 min

Doing a PhD in a Language You Can Barely Order Lunch In

Moving from Algeria to Tianjin, doing research in English, living in Mandarin, and thinking in French when nobody is looking.

open →
2026.04.24 research 5 min

How I Read a Paper Without Falling Asleep

You cannot read every paper, and you should not try. A three-pass method for deciding what deserves your afternoon.

open →
2026.04.17 models 6 min

Transformers Are Not Magic on Small Data

Vision Transformers are extraordinary given enough data. Most of us do not have enough data. Here is what actually happens.

open →
2026.04.10 models 6 min

How to Get Several Mediocre Models to Cooperate

Bagging, boosting, and stacking, explained without the textbook fog. Why ensembles win, and when they are not worth it.

open →
2026.04.03 method 5 min

Feature Scaling, and How to Not Shoot Yourself in the Foot

StandardScaler, MinMaxScaler, RobustScaler. Which one, when, and why outliers change the answer.

open →
2026.03.27 method 6 min

Nobody Warns You About Class Imbalance

SMOTE is not a magic spell, and applying it at the wrong moment will quietly ruin your evaluation. A practical guide.

open →
2026.03.20 method 5 min

Accuracy Is a Liar

A model that predicts 'healthy' for everyone can score 95%. Here is what to report instead, and why reviewers keep asking.

open →
2026.03.13 method 6 min

Your Model Is Probably Cheating

Data leakage is the reason your accuracy looks amazing and your reviewer looks unconvinced. Here is where it hides.

open →
2026.03.06 tools 7 min

Building Research Tracker: A Chrome Extension for Academic Researchers

A privacy-first extension that saves papers, manages deadlines, and recommends journals through the OpenAlex API. No server, no telemetry.

open →
2026.02.27 signal 6 min

What is MFCC and Why It Matters in Voice Analysis

The explanation I wish I'd had when I started: how Mel-Frequency Cepstral Coefficients work, and why they are still the workhorse of voice analysis.

open →
2026.02.20 personal 5 min

My Journey from Bioengineering to AI

How curiosity about the body led me to machine learning and a PhD in medical AI.

open →