Prakhar Kumar
Experienced in behavorial analytics, working with real time data. Worked on a diverse spectrum of problems such demand forecasting,sales propensity models, Optimisation, Impact assessment, Causal modelling, Supply chain analytics and credit and fraud modelling. Have a hands on experience working with Python and pyspark including the use of relevant libraries and packages as well as platforms such as AzureDB, HIVE and AWS Sagemaker. Have a working knowledge of deep learning models as well.
Graduated from IIT Roorkee in July 2019 with a Bachelors in Production and Industrial engineering.
REVIEWS FROM CLIENTS
Inventory planning and warehouse management -
• Developed a model to automate the process of manual supply chain planning and redu...
Inventory planning and warehouse management -
• Developed a model to automate the process of manual supply chain planning and reduced stockout by 30% for 1000+ SKUs across 5 warehouses by using sharper demand forecasts, relaible vendor TAT calculations and safety stock optimisations at a SKU X warehouse level.
Demand Forecasting -
• Developed a time series model to make gross demand forecasts at a city
supercategory level bringing down the error range from 25 to under 8%
• Model is an ensemble of SARIMAX and Fb-prophet that runs in an EC2
cluster every week via an airflow cronjob. The model has consistently
given an error(MAPE) of <10% for 13 supercategories across ~46+ cities
for the last 7 months.
• Increased accuracy resulted in a better partner onboarding plan,
inventory and CX/PX planning, resulting in a savings of 10cr in the month
of April 2023.
• Model results are currently consumed by finance team for CX/PX support
hiring, ops team for inventory management and planning as well as all
City teams across India to plan supply by onboarding partners.
Alerting System–
• Developed and deployed an unsupervised anomaly detection model to
track event metrics and raise alerts in cases of breakages of event pages.
• Used an EC2 server to deploy the model and it currently tracks ~100+
events in real time.
• Causal Modelling- Developed and deployed a CausalML based model to
understand the effect of various delivery treatments on the a...
• Causal Modelling- Developed and deployed a CausalML based model to
understand the effect of various delivery treatments on the account ACR
and get lifts and recommendations at an account level.
• Lead Prioritization- Developed a pipeline to rank best leads from 7
million yearly leads. The model is live and has resulted in 8 million dollars
of incremental revenue in 18 months.
• Impact measurement- Optimized the spend for multiple co-dependent
support services using Structured Equation Modelling by identifying their
independent impact on the overall sales of the drug.
• Launch level Sales forecasting- Conceptualized and developed a
predictive model to forecast the demand of a newly launched drug using
analogs from the same universe.
• Credit Score Modelling- Optimized and co-developed a custom credit
score model, for Indifi’s potential customers using CIBIL sco...
• Credit Score Modelling- Optimized and co-developed a custom credit
score model, for Indifi’s potential customers using CIBIL score and bank
details to determine loan amount qualification of a customer.
• Fraud detection module- Co-developed a Fraud engine algorithm using
alternate data from Truecaller and other messaging and email apps to
detect potential fraud applicants.