NIMEESH / DATA LAB Portfolio

SCIENTIFIC & FINANCIAL EXPLORATIONS

A PERSONAL PROJECT ARCHIVE

Data analysis.

Behind every dataset,
a question worth asking.

From distant planets to particle collisions and financial markets. I use code to look for patterns, test ideas and learn what the evidence can actually tell us.

Explore the work
04documented studies
and plenty still to learn
THE SHAPE OF A SIGNAL

Every point starts a question.

GENERATIVE STUDY / ILLUSTRATIVE DATA SURFACE
FOLLOWING THE DATA THROUGHNASA Exoplanet ArchiveCERN LHCb open dataMarkets Yahoo Finance

01 SELECTED WORK

Four investigations.
One curious mind.

The question, the method, the finding.
Every study opens into the full report and code.

ASTRONOMYSTUDY / 01

NASA EXOPLANET ARCHIVE

The planets we find.
The patterns we miss.

Transit vs radial velocity: comparing orbital periods and reported masses, while accounting for duplicate entries, missing measurements and detection bias.

MEDIAN ORBITAL PERIODdays
Transit8.027 d
Radial velocity307.88 d

Observed archive samples, shaped by how planets are detected.

  • Python
  • Pandas
  • Distributions
Read report : exoplanets (PDF, opens in a new tab) View code: exoplanets (opens in a new tab)
PARTICLE PHYSICSSTUDY / 02

CERN OPEN DATA · LHCB

Finding a peak
in the collisions.

My first ROOT-file investigation. Extracting reconstructed B± candidate masses and fitting a Gaussian peak with a constant background.

FITTED PEAK CENTREMeV/c²
5279.4≈ μ
Normalized Gaussian peak componentModel curve reconstructed using the report's centre of approximately 5279.4 MeV per c squared and width of 17.75 MeV per c squared. These are model values, not observed candidate counts.

Normalized model component. A learning fit, not a precision mass measurement.

  • ROOT data
  • SciPy
  • Curve fitting
Read report : B meson mass peak (PDF, opens in a new tab) View code: CERN analysis (opens in a new tab)
FINANCIAL DATASTUDY / 03

UBS × USD/CHF · 2021–2026

An idea the data
didn't support.

Do UBS stock and USD/CHF converge? Five years of daily data showed near-zero linear return correlation and a trending relative-performance series.

DAILY RETURN CORRELATIONPearson r
−0.01approximately

1,253 common daily observations. Correlation alone does not test convergence.

  • yfinance
  • Time series
  • Correlation
Read report : UBS and USD/CHF (PDF, opens in a new tab) View code: UBS and USD/CHF (opens in a new tab)
PARTICLE PHYSICSSTUDY / 04

CERN OPEN DATA · THE FOLLOW-UP

Beyond the peak.
Into the assumptions.

Testing vertex-quality selections, comparing signal and background models, and using residuals to see where a fit still falls short.

CANDIDATES RETAINED AFTER SELECTION%
χ²/NDOF < 587.33%
χ²/NDOF < 380.07%
χ²/NDOF < 269.91%

Tighter cuts retain fewer candidates. Retention is not a measure of signal purity.

  • Selection cuts
  • Model comparison
  • Residuals
Read report : candidate selection and models (PDF, opens in a new tab) View code: CERN follow-up (opens in a new tab)

PROJECT SNAPSHOTS VALUES FROM THE LINKED REPORTS. METHODS AND LIMITATIONS INSIDE.

02 THE APPROACH

The plot is only
the beginning.

I'm a physics undergraduate learning by working with real datasets. I document the choices, mistakes and limitations as carefully as the results. Each investigation is another step toward better scientific judgement.

MY WORKING TOOLKIT

Python / Pandas / NumPy
Matplotlib / SciPy / ROOT files

  1. 01

    Start with a question.

    Make the idea specific enough that data could change my mind.

  2. 02

    Know what a row means.

    Check provenance, duplicates, missing values and selection effects before plotting.

  3. 03

    Test the interpretation.

    Compare models, inspect residuals and look for assumptions that don't hold.

  4. 04

    Leave a clear trail.

    Publish the report and code, including what the analysis does not establish.