Nov 2023 – Sep 2024 · 10 mos · Bengaluru, Karnataka, India · Hybrid

• Financial anomaly detection on millions of transactions;
• Schedule optimiser for Virtual Expert platform

Jul 2022 – Nov 2023 · 1 yr 4 mos · Bengaluru, Karnataka, India

• Large scale demand forecasting for Swiggy Instamart across multiple categories
• Restaurant Ads recommendation in Swiggy Food

Jun 2020 – Jul 2022 · 2 yrs 1 mo · Bengaluru Area, India

• Enabled business to use ML-driven forecasts for labour, resource and operations planning for two consecutive fiscal years:
• Developed a resusable framework to generate ML-driven forecasts at scale for a large number (approx. 200,000) of univariate time sereis
• Implemented a set of novel techniques in PySpark, e.g., RLM-based outlier treatment for time sereis; extension of univariate time sereis modeling for a cluster of sereis; temporal cross-validation based model selection; auto-regressive LASSO etc.
• Drove the Data Sciene project end-to-end; Collaborated with team members and mentored junior employees and interns
• Helped business to scale the ingestion of new ideas provided by the associates to improve efficacy of several business units:
• Experimented with multiple DL-based text-classification models to map ideas to business units
• Improved the model by introducing an iterative approach of model-based soft-labeling of ideas

Jan 2020 – May 2020 · 4 mos

• Telecom based automotive solutions using GPS and inertial mobile sensors
• Built a hyper personalised prediction mechanism for seggregating passanger trips from driver trips which helped Zendrive to win an important pilot
• Created an end-to-end framework to classify trips as motorbike vs. car. Worked with the engineering team to deploy the model in production

Jan 2019 – Jan 2020 · 1 yr

• Telematics insurance risk modeling using mobile sensor data
• Collaborated with the team members to build GLM based insurance risk model which achieved 8x lift. This model has been used by several Zendrive customers for insurance pricings and discounts
• Single-handedly created an end-to-end ETL framework in PySpark to generate aggregated driving behavior data (more than 10 billion miles) to be used in modeling

Jul 2018 – Jan 2019 · 6 mos

• Sensor fusion based event detection
• Built a generic peak detector algorithm for sensor fused signals which has been used in production to detect events like hard brake, rapid acceleration, hard turns etc. These events are used by downstream models for driver scoring, insurance risk scoring.

May 2017 – Jul 2017 · 2 mos

• Implemented a new algorithm to find the parameter estimates in logistic regression. Also I considered the case with missing data in binary outcome set up and performed a detailed comparison study of various available multiple imputation methods.