Designed business proposals and collaborated with clients to prepare Analytical Data Sets (ADS), perform exploratory data analysis (EDA), fine-tune regression models, simulate use-case scenarios, optimize budgets and targets, and deliver post-model analytics and presentations.Built Marketing Mix Models (MMM) using regression techniques and optimization methods, driving efficiency and ROI for marketing campaigns.Developed a custom optimization algorithm to allocate budgets and KPIs across diverse media channels, improving resource utilization and campaign effectiveness.Engineered a simulator tool for MMM, enabling scenario simulations and actionable insights into potential outcomes.Established robust data pipelines for media streams, ensuring accurate and timely data integration.Leveraged AWS SageMaker, Athena, Python, SQL, and Postman to develop tools and pipelines essential for analytical projects.Contributed to the Pricing and Vehicle Incentive Steering Analytics project by creating an incentive optimization model and coordinating with field teams for pilot runs.Applied advanced data preparation techniques using PySpark, creating comprehensive datasets and implementing a four-stage modeling approach for discount steering, leading to improved sales and incentive optimization.Worked across domains including Automotive, Healthcare, Finance, Retail, and Supply Chain, collaborating with cross-functional teams to implement analytical models at scale.Utilized cloud computing (AWS EC2), PySpark, and advanced analytics tools like SQL and Excel to execute large-scale, data-driven solutions.

Client: MRO Optimization for Chemical Manufacturer in Saudi ArabiaThe Client would like to build a reliable AI-driven framework to have efficient inventory management make duplicate detection more effective and maximize the approval rate and extent of the solution to help visualize the output.Kind of Data: The client had historical data of more than 2 million spare parts in their MRO supply have a greater number of duplicated spare parts in inventory.Data processing: Data is captured and stored in the Pi-log server and extracted to Hive storage. Creating structured data, cleaning the text, characteristics, values, extraction of design number, manufacturer number, and data enrichment for missing values.Problem type: Inventory management of MRO, identifying duplicates is crucial for effective inventory management. Outcome: The client will be able to access the multiple dashboards to help in the accountability and governance structure of the spare parts maintenance, user-friendly search portal that enables full-text search autosuggestions helps in cost-saving and solution increased the duplicate approval rate by 40%.Stacks used:Python, Pyspark, SQL Dataiku, Rapid Miner, Azure DevOps, SonarQube, Hue.Client: Largest technology supplier for Food processing and other industries, GermanyProject: To build a framework to forecast any business metrics to increase the quality of sales. Problem type: Time series problem to forecast different combinations of business units.Approach: Built 15 different time series algorithms; Chosen the best model with the least errors. Outcome: Client business will be able to forecast different business units with utmost accuracy from 15 different models. Sales and Finance teams can use these forecast results to plan the business for future time periods. Tools used: Azure SQL database, Azure Data bricks, Azure data studio, Azure Data Factory.

Client: Petro Chemical Manufacturer in Saudi ArabiaThe client faces different slippage incidents or outages in many Towers of the manufacturing plants. The incident leads to the shutdown of equipment or an entire tower or part of the plant. It causes financial loss due to the off-specification of output from the Tower. The client wanted an early warning system to predict the slippage incident using past data.Project: To predict the incidents early to avoid outage or shutdown of the equipment or entire Tower. Kind of Data: IoT sensors collect data about the chemical flow, level, temperature, pressure, and many metrics.Data Extraction: Data is captured every second; we aggregated data at a minute level.Data processing: Handled missing data, dropped redundant variables, added different date variables, and lag variables.Problem type: Regression problem to predict Incident (Anomaly incident) using past data. Approach: Built XG Boost model for good accuracy and built Ridge regression as an explanatory model.Outcome: The plant operator will get an alert in the dashboard about the incident 4 hours in advance and 1 hour in advance.The operator can adjust the input feed using the influencing variables from the explanatory model.Use-cases worked on:• Predicted Slippage of Acid gases (CO2 and H2S) in Caustic Tower/Cleaning Tower.• Predicted Stack Temperature in Furnace Tower/Heating Tower. • Predicted Overhead temperature in Quench Tower/Cooling Tower• End to End Data science Life cycle.• Requirement Gathering, Problem understanding, Data extraction, Data preprocessing, Data modeling, Model Hyperparameter tuning, and Model deployment.