Data Science Lead · Bengaluru, India

Ayush Singh

I build predictive models, forecasting systems, and Gen-AI applications that move retail, supply-chain, and logistics numbers. 6+ years turning noisy data into decisions — $1.3M in inventory savings, 12% forecast lift, and an ROC-AUC of 0.92 on returns, to name a few.

Try the tools ↓ LinkedIn ↗ Email
6+
Years in data science
$1.3M
Inventory savings delivered
12%
Forecast accuracy lift
0.92
ROC-AUC on returns model
01 / About

Analytics, but pointed at the P&L.

I lead data science at Impact Analytics, where I designed a Merchandise Financial Planning solution and the Airflow pipeline behind it for a major North American retailer. Before that, I spent five years at Coupa (née Llamasoft) shipping forecasting and supply-chain design models to customers worth $500K–$700K in portfolio value.

My favourite problems sit at the intersection of classical ML, time-series forecasting, and more recently RAG-based Gen-AI. I like models that earn their keep in dollars, not dashboards.

What I'm doing now

  • Leading a team of 4 on MFP for a NA retail brand
  • Cluster-specific Random Forest sales forecasters for 250+ stores
  • A/B-tested markdown strategies → +8% margin
  • Mentoring 100+ aspiring data scientists at AlmaBetter
02 / Skills

The stack I reach for.

Languages & Tools

Python SQL R Google BigQuery GCP Tableau Airflow Git

Machine Learning

Scikit-Learn TensorFlow Pandas Random Forest XGBoost KNN K-Means Logistic Regression Deep Learning NLP Topic Modeling

Techniques

Time-Series Forecasting Predictive Analytics A/B Testing Optimization Cross-Validation Feature Engineering

Gen-AI

LangChain pgvector RAG applications
03 / Experience

What I've shipped.

Data Science Lead

Impact Analytics · Bengaluru, India · Jun 2024 — Present
  • Led a team of four to stand up an Airflow ingestion pipeline and implement a Merchandise Financial Planning solution for a major NA retail brand.
  • Segmented 250+ retail stores with K-Means and built cluster-specific Random Forest forecasters with time-aware features — +12% forecast accuracy.
  • Optimised markdown strategy via A/B testing on pricing approaches → +8% margin for a NA clothing retailer.
  • Deployed an XGBoost product-returns model that beat KNN, RF, and logistic regression (ROC-AUC 0.92) — saving ~$450K/yr in return costs.

Senior Data Scientist — Customer Adoption

Coupa Software · Pune (Remote) · Mar 2022 — Sep 2023
  • Managed a $500K–$700K client portfolio, promoting Coupa's Supply Chain Design product suite.
  • Built and deployed a RAG-based Q&A chatbot on LangChain for support FAQs, integrated into in-house SaaS apps.
  • Forecast monthly demand for 3,000+ spare parts with ARIMA / SARIMA / SARIMAX (wMAPE 9.6%) — $1.3M inventory savings for an automotive giant.

Data Scientist — Customer Adoption

Coupa Software (acq. Llamasoft) · Pune (Remote) · Oct 2021 — Mar 2022
  • Built a cloud-based planning-and-approval workflow app on Coupa App Studio, later productised as an off-the-shelf offering.

Data Scientist

Opex Analytics (a Llamasoft company) · Pune (Remote) · Aug 2018 — Oct 2021
  • Cut runtime and memory of ingestion and reporting scripts by 20–30% using memory profiling.
  • Designed Tableau dashboards for Fill Rate, Dispatch Rate, Sales, and Cost KPIs.
  • Shipped an ML solution (Decision Tree, Random Forest) predicting safety-incident likelihood for a railroad operator.
  • Led a team of 3 across 8 accounts covering UAT, IP maintenance, and enhancements.
04 / Interactive Tools

Play with the techniques behind the résumé.

Each tool is a tiny, client-side reproduction of something I've shipped at work — no servers, no tracking, just browser math. Open the dev tools if you want to read the source.

📈

Sales Forecast Lab

Holt-Winters exponential smoothing on a retail sales series. Tweak level, trend, and seasonality to see the forecast move live.

time-seriesforecastingHolt-Winters
🧪

A/B Test Calculator

Two-proportion z-test with p-value, confidence interval, and lift. The same math I used to green-light markdown strategies.

statisticshypothesis-testing
🎯

K-Means Store Clustering

Click to drop "store" points on the canvas, pick k, and watch Lloyd's algorithm iterate to convergence.

unsupervisedsegmentationLloyd's
📦

Return Risk Predictor

A logistic-regression-style scorer for "will this order be returned?" with ROC curve and threshold tuning.

classificationROC-AUC
05 / Education

Education

M.Sc. Operational Research

Kirori Mal College, University of Delhi · 2018

B.Sc. Statistics

Patna Science College · 2016

06 / Awards

Awards & Recognition

Gen-AI Adoption Award

May 2025 · Driving business impact through AI adoption

Spot Award (×2)

Feb 2020 & Mar 2022 · Customer success through collaboration

Data Science Mentor — AlmaBetter

Sep 2021 — Present · Mentored 100+ aspiring data scientists

Let's talk data.

Roles, collaborations, a stats question at 2am — whichever. I reply.

ayusharyan45@gmail.com LinkedIn ↗ +91 95796 96071