Anuj NagarData analysis & science writing

I turn messy machine data into clear answers.

Data analysis in Python, SQL and Power BI. And science explained in plain words, at UnFogIt.

See the flagship project Read the field notes

Signal lab

synthetic data
A starts B starts PATH APATH B
Raw timestamps: B vs A +2.0 msOKNo offset removed yet

Looks fine: A seems to start first.

A replay of the demo recording in my MF4 project: the raw view hides a violation.

About the notebook

I’m Anuj. For five years I’ve checked whether automotive software behaves, chasing millisecond start-up bugs through gigabytes of measurement logs.

Now I’m turning that habit into data analysis, and I write UnFogIt, where research too dense for headlines gets explained in plain words.

Python repopandas repoNumPy repoasammdf repoNLP repoPower BI repoDAX repoExcel repopytest repoGitHub Actions repoD3.js repoSQL courseTableau course

Every skill links to the repo that shows it. If I can’t point at the work, it isn’t on this list.

Experiment 01 · flagship

MF4 Log Analytics

Two processors, two clocks, hundreds of power-ups. For every start-up in a multi-gigabyte recording, which one came first, by how many milliseconds, and why.

95 h 2.5 h

From about 95 hours to about 2.5 hours. Estimated for a 10-recording campaign (about 585 start-ups). The tool time was measured once; the manual time is my estimate of 10 minutes per cycle.

  1. 01LoadOpen multi-GB MF4 files lazily
  2. 02SegmentSplit into power-ups
  3. 03MeasureResets, gaps, first app update
  4. 04CorrectRemove clock offset between paths
  5. 05Cross-checkA second, offset-free method
  6. 06JudgeOK, NOK, CHECK or INFO
  7. 07ReportCSV plus browser dashboard
Read the case study Open the live workflow Source on GitHub

More experiments

The rest of the bench

All experiments
02

ChatGPT Review Analysis

What do 193,148 app reviews actually say? Python · TextBlob · NLP · Jupyter
Oct 2026 03

Global IT Spend in Power BI

Where does a $900M plan drift from the forecast and the actuals? Power BI · DAX · Power Query · Python EDA
Oct 2026 04

Coca-Cola US Sales in Excel

Selling more is not the same as earning more. Excel · SUMIFS · Dashboards
Oct 2026 05

Python Test Automation Framework

CAN tests that run, report and publish themselves on every push. pytest · Allure · GitHub Actions · python-can
Mar 2026 06

Python DSA Learning Log

26 practice files and 9 interactive note pages, in public. Python · Algorithms · Data structures
Mar to May 2026

Field notes · UnFogIt

Wait, what?

UnFogIt breaks down the AI, science and tech research too dense for headlines.

All 12 notes
Biology

The Lilliputian Molecule

How DNA Sequencing Solved a 90-Year-Old Mystery of Mushrooms That Make People See Tiny Humans

 Read on UnFogIt Space

The Triple Engine

How Webb Caught Three Feeding Supermassive Black Holes Colliding in a Single Dawn-Era Galaxy

 Read on UnFogIt AI

The Softmax Fix

How Google's Jonathan Barron Solved a Decade-Old Floating-Point Bug in Deep Learning's Core Engine

 Read on UnFogIt Space

The Magnetic Tapestry

How Sunrise III's Stratospheric Balloon Telescope Uncovered the Twisted 3D Web Powering the "Quiet" Sun

 Read on UnFogIt

Lab log

What I learned, and when

Full log
  1. built

    MF4 Log Analytics goes public

    Open-sourced the pipeline and dashboard, with a synthetic demo recording so anyone can run it, plus an interactive workflow page. See more →

  2. built

    Global IT spend dashboard in Power BI

    Plan versus forecast versus actual on 93,729 rows, with DAX measures, Power Query prep and a Python EDA notebook. See more →

  3. wrote

    Twelve new UnFogIt explainers

    Between 23 and 26 October: dark energy, quantum error correction, plant biosensors and more. Each takes a paper too dense for headlines and says what it found, in plain words. See more →

  4. built

    ChatGPT review analysis

    Sentiment and aspect analysis on 193,148 app reviews. Learned the hard way that star ratings and review text disagree more than you would expect. See more →

  5. built

    Coca-Cola sales dashboard in Excel

    SUMIFS, heatmaps and a dropdown explorer. The biggest market was not the most profitable one. See more →

  6. learned

    Power BI and Tableau

    Crio.Do course completed.

On the bench

What I’m building next

In progress

Finish the Crio.Do DA-201 capstone

The data analytics programme has one big project left. It goes on this site as entry 07 when it is done.

Plan

A second analytics project on public telemetry data

Same habits as the MF4 work (two methods, a clear verdict, honest caveats), but on data anyone can download and check.

Plan

A chart in every UnFogIt piece

Pull one number out of each paper and draw it, so a reader can see the claim as well as read it.

Plans, not promises. Dashed cards are ideas; the solid one is already under way.

Contact

Got a messy dataset or a science question?

I read everything and answer when I can. The quickest way is email, and LinkedIn works too.

Copy my email LinkedIn GitHub UnFogIt