Sumit Singh

About the Candidate

1. Senior Data Scientist, Remote Job,Nestle, Switzerland. Aug 2019 – till date
LSTM, VAR & VECM implementation for Price Prediction of commodities futures and energy prices.
GBM & design of ensemble for selection from different models.
Impact: The prediction beats the benchmark 70%+ times as shown by back testing.
Coding and implementation of Mean Reversion for Price Risk Management for commodities futures in
Python. Implementing it for energy as well.
Impact: Back Testing shows savings around 100m CHF per year.

2. Senior Data Scientist, Telecom Analytics, Subex, Bangalore. Jun 2017- Jul 2019
Worked as a senior data scientist to create statistical modeling with a team. Main projects: Anomaly Analytics and Social Network Analytics using large scale telecom data. Used algorithms/ methods such as PCA, LSTM with Time Series to detect anomaly and minimize false positives.
Worked on diffusion modelling to create ideas for Social/Cell network analysis using networkx, pandas package in Python.
Project management work such as planning KRAs. Designing and conducting training on Statistical Learning, R and Python.

3. Manager, Fin Enterprise Analytics Philips Innovation Campus, Bangalore Jan2016-May 2017
Projects: Working Capital Analytics, Factory P&L Analysis. Methods Used: ML(Decision Tree, Random Forest,
PCA, Cluster Analysis), ARIMA. Impact – 10% reduction in locked working capital. Languages: SQl, R, Python.

4. PhD Paper3

Inventory Control of Perishable Inventory which can be converted into different forms
Desc: Multi-Period decision problem for convertible perishable products with stochastic demand (with different statistical distributions) and price in different periods. Techniques Used: Convex Optimization, News-vendor Model, Differential Calculus, Sensitivity analysis, Simulation in Python.

5. PhD Paper 1 and 2
Inventory Control of Two Stage Perishable Inventory
Desc: Stochastic Deterministic model for two stage perishable inventory (used in Airline, Auto mobile and PCB industry) control.
Techniques used: EOQ Model, discreet time Markov chain, Queuing Theory, Alternating Renewal Process and various statistical distributions.
Result: Closed form/Algorithmic solution to determine lot size, unit size and N policy of queuing.
Techniques Explored: Markov Decision Process, Multi-armed Bandit models, Dynamic Programming, Stochastic Programming, Other Machine Learning Algorithms.