Yash Desai

• Built an end-to-end BTL campaign management platform with automated customer segmentation,
cohort selection, and incremental revenue analysis, enabling marketing teams to optimize budget
allocation across 24 brands.
• Operationalized a customer churn prediction system using ensemble modelling (Logistic
Regression, Random Forest, LightGBM), achieving ~91% churn recall and directly enabling
targeted retention campaigns that improved customer lifetime value.
• Designed and deployed a demand forecasting engine using ensemble ML models, achieving on an
average 23% improvement in forecast accuracy and reducing manual planning effort significantly.
• Implemented a centralized allocation and replenishment engine combining demand forecasts with
constraint-based optimization, resulting in 20% improvement in product availability and 70%
reduction in sales loss.
• Developed an LLM-powered automated insights engine that analyzes KPIs, generates narrative
reports, and sends anomaly-based alerts to stakeholders via scheduled email notifications.
• Built interactive Streamlit dashboards for business performance tracking and executive summary
views consumed by senior leadership.

• Owned the complete forecasting pipeline from data prep to deployment, delivering 5-10%
improvement in forecast accuracy by refining statistical feature engineering and ensembling
conventional time-series models with ML algorithms.
• Developed demand forecasting logic as reusable platform plugins supporting parallel ML and
statistical model execution at DFU level, enabling large-scale enterprise S&OP and demand
planning across multiple clients.

• Implemented a pull-based dispatch optimization engine prioritizing PO fulfillment using lead-time
and urgency constraints while optimizing vehicle mix and capacity utilization, saving 3 hours of
daily manual planning and improving service levels by 3-4%.
• Built an analytical decision model to evaluate and optimize last-mile carrier selection by
benchmarking FTL and PTL rates across vehicle categories and distance bands, identifying cost
optimal delivery options at regional level.
• Developed a demand forecasting framework using ensemble learning (XGBoost, LightGBM,
ARIMA), which led to 7-20% improvement in forecasting accuracy across FMCG and FMCD use
cases while mitigating individual model bias and variance.
• Designed an outlier correction framework leveraging time-series decomposition and residual
pattern analysis, resulting in 2-5% improvement in forecast accuracy across multiple projects.