Amit Pathak

Disruption Recovery Solver for Supply Chain Disruptions

• The project aims to provide organizations with a powerful tool to proactively manage and recover from supply chain disruptions.
• By optimizing recovery strategies using OR methods, businesses can reduce downtime, minimize financial losses, and maintain customer satisfaction even in the face of unforeseen disruptions.
• Create a solver that generates optimal recovery strategies in the event of a disruption, taking into account factors like inventory management, transportation rerouting, and resource allocation.

Inventory Optimization using OR tools

• The project improves inventory management practices, resulting in cost savings and improved operational efficiency.
• The project combines a wide array of OR tools and techniques, such as mathematical modeling, optimization algorithms, simulation, and data analytics, to achieve the objectives.
• The objectives include reducing the risk of overstocking and understocking, order aggregation from multiple vendors. Estimation of order delivery, forecasting orders, identifying best warehouse routes, recommending best-fitting package, etc.

Digitalizing Contracts & Services Procurement

Sentence Transformer

Rapidminer

Python

• As part of digitizing contracts and services procurement, an AI search functionality is built to leverage the power of sentence transformer models to provide desired responses for a user search made for a contract or service.
• The system is built and deployed on RapidMiner platform.

Vendor Procurement Analysis & Vendor Recommendation Engine

MLOps

XGBoost

LightGBM

Statistical Models

• Implemented end-to-end software solutions on the principles of MLOps lifecycle management using Databricks MLFlow library.
• Analysis of the vendors based on 7 different metrics including popularity of a vendor, delivery duration, pricing of items, market reputation, etc. Each vendor is compared with their counterparts based on the different combination of items present with them.
• The vendors were analyzed and ranked accordingly. The estimated delivery time of the vendor is modelled using the XGBoost and LightGBM models whereas other metrics are built on NLP engine for market reputation and heuristic models.

Discount Recommendation for Interest based Payout for the Business Sellers

Linear Regression

Statistical Model

Clustering

Docker

• Built a system to provide dynamic interest rate for the early payment that can be made to a given business seller based on its firmographics and the type of historical business provided by the Seller.
• This helps the businesses having ample amount of cashflow save huge on operational costs and have a reward mechanism to encourage maximum seller participation.
• The system is built using SVM's clustering approach on top of a statistical model which calculates 16 different Seller metrics using the historical time-bound data for each of the Seller.
• The system is further enhanced to predict the acceptance of a Seller for a given interest-based payout by the businesses to maximize the participation of them by using the regression based linear models.
• The system is designed end-to-end and deployed using docker containers to the target environment.

Spare Parts Forecasting to improve Supply Chain Efficiency

Survival Analysis

Forecasting

• Design a Survival model to learn the time horizon for the breakdown of appliances and forecast the requirement of spare parts for the next six months.
• The final solution is implemented with the help of four different models whose input and (or) output are inter-connected.
• The first model is a standard ARIMA model to forecast the primary sales of appliances followed by an algorithm built using cohort analysis to forecast the secondary sales. The third model predicts the spare parts requirements by using the survival model - Cox Proportionality Hazard. The final model is a feedback model which simulates the existing inventory, their location and transit times and accordingly modify the final requirement of spare parts that needs to be ordered/stored/manufactured.

Timeseries forecasting for Metal Procurement

Timeseries Forecasting

ARIMAX

SARIMAX

• Enhance the existing timeseries VAR models to a neural network-based approach to achieve higher accuracy.
• Maintain the current models by regularly updating data and analyzing the changing trends and share report with the client team to help them take better procurement decisions.
• Automated the complete data download and processing stage to reduce the manual effort from 6 hours to a few seconds.
• Refactoring the existing code, building libraries and modules to make the usage of the models more efficient.

Article & News Recommendation Model using Heuristics Approach

Heuristics Modeling

NLP

Recommendation Engine

Docker

• Build a heuristics-based model to provide dynamic scoring to news and articles based on the user's profile who has logged in.
• The scores are developed using a ranking-based algorithm for the articles and highest-score for the news by considering the oldness, category, sub-category, business vertical, user horizontal, word semantics, etc. for each of the news and articles on the platform.
• Designed the complete end to end solution and deployed using docker containers to the target environment.

Intruder Detection System GUI Application - Security Surveillance

GUI Programming

Computer Vision

PyQt5

Security

• Developed a PyQt5 desktop application for Intruder detection surveillance using OpenCV.
• The application captures a series of images if any movement is observed in a surveillance room during a given timeframe and sends a GIF to the concerned authorized person to take necessary security measure if required.