Data Scientist — Hyderabad, IN — Open to Relocate

I turn messy, real data
into decisions worth trusting.

BBA-trained analyst who moved into machine learning — Python, Scikit-learn, and statistical rigor applied to real datasets, end to end: cleaning, modeling, validation, and honest reporting of what the numbers actually support.

View projects Get in touch

About

I came into data science through business analytics — two internships spent building Power BI dashboards and finding the operational insight buried in messy multi-source data. That habit of asking "does this number actually mean what it looks like it means" is what I now bring to machine learning: every project below states its limitations as plainly as its results.

  • Based inHyderabad, India
  • EducationBBA, 8.64 GPA
  • Core stackPython · scikit-learn
  • AlsoSQL · Power BI
  • RelocationOpen, anywhere

Selected work

Five projects, five real datasets

01
Flagship

Bitcoin Trader Sentiment Analysis

Investigated whether Bitcoin market sentiment (Fear/Greed Index) predicts individual trader outcomes, using 104K+ real closed trades from Hyperliquid matched against 2,644 days of sentiment data. Traced win-rate differences down to whether they were whale-driven or broad-based before trusting any conclusion, then quantified sentiment's predictive power with a cross-validated classifier.

Limitation acknowledged: only 32 trader accounts — cluster-level conclusions are directional, not statistically conclusive. Correlation, not proven causation.

View code →
0.85
ROC-AUC · 5-fold CV
89.2%
Win rate, Extreme Greed
p<.001
Two-proportion z-test
02
End-to-end pipeline

Hotel Revenue & Cancellation Prediction

Built two supervised models on 87K+ hotel bookings sharing one leakage-safe preprocessing pipeline: a regression model predicting nightly room rate (ADR), and a classifier predicting booking cancellation. Tuned via grid search, evaluated on a held-out stratified test set.

Limitation acknowledged: cancellation model recalls 61% of true cancellations at 75% precision — a real trade-off from class imbalance, disclosed rather than hidden behind aggregate accuracy.

View code →
0.88
R² · ADR regression
0.87
ROC-AUC · cancellation
87K+
Bookings analyzed
03
Climate & agri-risk

Global Crop Disease Risk Prediction

End-to-end pipeline on a 120,000-row, 43-feature climate and agriculture dataset — correlation analysis across climate, soil, and pest variables, then a Random Forest regressor predicting expected crop yield loss, plus a severity classifier for early-warning triage by country and crop type.

Limitation acknowledged: risk scores reflect historical climate patterns; not validated against real-world outbreak data outside this dataset's window.

View code →
0.69
R² · yield loss
3.6
MAE, unseen data
120K+
Rows, 43 features
04
Computer vision

Microscopy Blood Cell Classification

CNN classifier on 17K+ microscopy images, built on MobileNetV2 transfer learning. Replaced the standard Flatten head with GlobalAveragePooling2D — cutting head parameters roughly 250× and removing the overfitting that came with the larger head — and handled severe class imbalance with weighted training rather than naive oversampling.

Limitation acknowledged: trained on a single public microscopy dataset; would need validation against a different lab's imaging setup before any clinical-adjacent use.

View code →
17K+
Microscopy images
250×
Head param reduction
MobileNetV2
Transfer learning base
05
Preprocessing pipeline

Customer Churn Prediction

Designed a full, reproducible preprocessing pipeline for a mixed-type churn dataset: mean/median/most-frequent and KNN imputation matched to each field's missingness pattern, One-Hot and Ordinal encoding via ColumnTransformer, and StandardScaler feature scaling — wrapped into a single sklearn Pipeline so training and inference never drift apart.

Limitation acknowledged: project scope centers on pipeline design and reproducibility; model comparison across KNN/Decision Tree/Logistic Regression is a smaller, secondary part of the work.

View code →
1
Unified sklearn Pipeline
4
Imputation strategies compared

Toolbox

What I actually work with

Data Science

  • Python · Pandas · NumPy
  • Scikit-learn
  • Feature engineering & scaling
  • Cross-validation, model evaluation

Deep Learning

  • Convolutional Neural Networks
  • Transfer learning (MobileNetV2)
  • ANN, RNN, LSTM fundamentals
  • OpenCV

Analytics & BI

  • SQL
  • Power BI dashboards
  • Advanced Excel (VBA, Power Query)
  • Statistical analysis, EDA

Background

Experience & education

Apr '25 — Oct '25

Business Associate

Asynzx

Analyzed multi-regional operational data end-to-end, improving key performance metrics by 15%. Built Power BI KPI dashboards that exposed workflow inefficiencies and reduced process bottlenecks by 20%.

Sep '24 — Nov '24

Analyst Trainee Intern

Forum Business Pvt. Ltd

Engineered interactive Power BI dashboards and automated reporting pipelines, improving reporting efficiency by 25%. Aggregated multi-source data to surface performance patterns.

Oct '22 — Apr '25

Bachelor of Business Administration

Avinash College of Commerce · GPA 8.64/10

Hyderabad, India.

2025 — 2026

Certifications

Simplilearn + Microsoft

Machine Learning Using Python (Jul '26) · Certified Business Analyst Program, CBAP (Sep '25)