Rupesh Singh Rawat
Overview
Technical skills
Timeline
- Built a customer booking conversion classifier utilizing Random Forest, evaluating predictive performance with ROC-AUC metrics to flag high-intent travelers.
• Constructed a lounge demand forecasting model analyzing haul type and time-of- day traffic patterns to optimize Heathrow Terminal 3 passenger flow.
- Worked on a data science job simulation focused on identifying high-intent travelers using engagement metrics.
- Analyzed passenger flow patterns related to Terminal 3 and supported the scenario’s analytical tasks. Produced insights based on the provided simulation data.
• Conducted exploratory data analysis (EDA) and data cleansing across multi-region retail datasets to identify key revenue drivers and underperforming categories.
• Designed executive-facing Power BI dashboards with custom DAX KPIs and interactive filters aligned with business stakeholder reporting needs.
📌 SITUATION
Analyzed a telecom dataset of 6,418 customers where 28.8%
churn rate was detected, putting ₹34.1L in annual revenue
at risk. The core business challenge was identifying why
customers were leaving and predicting who would churn next
— before it happened.
📌 TASK
Designed and delivered a complete end-to-end data analytics
solution covering data engineering, machine learning
prediction, and executive-level BI reporting — independently,
from raw data to actionable dashboard.
📌 ACTION
▸ Built a SQL Server data pipeline — ingested raw data into
a staging table, performed NULL analysis across 32 columns,
applied business-logic-based cleaning (ISNULL strategy),
and loaded production-ready data into prod_Churn table
▸ Created 2 SQL Views (vw_ChurnData, vw_JoinData) to
separate existing vs new customers for Power BI integration
▸ Trained a Random Forest Classifier in Python achieving 83%
accuracy and 85% Cross-Validation score (±0.01 std) —
handled class imbalance (847 Stayed vs 355 Churned) using
class_weight='balanced' parameter
▸ Identified Contract Type as the #1 churn driver (15%
feature importance) — validated by both EDA and ML model
▸ Scored 411 new customers with Churn_Probability_% —
predicted 368 at-risk customers, segmented into 4 risk
tiers: Critical (90-100%), High (75-89%), Medium (60-74%),
Moderate (50-59%)
▸ Designed a 2-page interactive Power BI dashboard with KPI
cards, risk category charts, geographic distribution across
22 Indian states, and a priority customer-at-risk table
📌 RESULT
▸ 236 customers flagged as Critical risk (90-100% churn
probability) with avg churn score of 89.66% — enabling
sales team to take immediate retention action
SITUATION:
A global e-commerce company (DataCo) was suffering
from a systemic delivery crisis — 54.71% of all
orders were delivered late across 6 international
regions, putting $2.1M in profit at risk from
94,523 delayed orders alone.
TASK:
As the sole analyst, I was responsible for
diagnosing the root cause of this crisis,
quantifying its financial impact, building a
predictive model to flag at-risk orders before
dispatch, and delivering an executive dashboard
for real-time decision-support.
ACTION:
▸ Analysed 172,765 order records (2015–2018)
using Python (Pandas, NumPy) after cleaning
raw data from 180,519 rows and 53 columns
▸ Engineered 6 new analytical features including
Order Processing Time, Delay, Is_Delayed, and
Profitability Flag to enable downstream analysis
▸ Built a custom Bottleneck Detection Framework
across 6 categorical dimensions — identifying
First Class Shipping at 100% delay rate as the
primary root cause (confirmed by 3 independent
methods: manual analysis, regional RCA, and ML
feature importance)
▸ Trained a Random Forest Classifier with SMOTE
oversampling — achieving 74% accuracy,
0.78 precision, and 0.75 recall on 34,553
unseen test records
▸ Designed a 2-page interactive Power BI dashboard
with 5 dynamic slicers, 6 KPI cards, Key
Influencers visual, DAX measures, and
conditional formatting
RESULT:
▸ Delivered a 3-tier prioritised recommendation
framework targeting reduction of late delivery
rate from 54.71% → below 30% within 12 months
▸ Estimated recovery of $1.2M–$1.5M in at-risk
profit through shipping mode rationalisation
and predictive alert deployment
- Extend the analytics work into an explicit ML workflow: add a training script, clear train/val/test splits, evaluation metrics, and a reproducible experiment log (W&B or MLflow).
- Harden data engineering and deployment: remove hardcoded connection strings, add secrets handling and a simple CI pipeline with unit tests for core data functions like get_efficiency_report.
- Productize the system by separating business logic from UI, adding model serialization and an inference API or lightweight service (e.g., FastAPI) for programmatic access to predictions and scoring.
- Use to build interactive analytics dashboards and data exploration tools for sports or similar domains using Streamlit and pandas.
- Develop the backend data access layer into a production-ready service - add parameterized queries, environment-based secrets, schema migrations, and connection pooling.
- Harden production readiness by adding structured logging, unit tests, retry/backoff for transient DB failures, and CI/CD pipelines.
Pandas
- Develop interactive analytics dashboards and data-driven MVPs for sports or small-scale operational use cases.
- Improve data engineering hygiene by adding environment secret management, unit tests for data transformations, and simple CI pipelines.
- Expand analytical rigor by adding clear evaluation protocols, baseline comparisons, and uncertainty reporting for any scoring or prediction logic.