Jul 2024 – Apr 2025 · 9 mos · Gurugram, Haryana, India

• Spearheaded the logistics and consumer product strategy for Zomato Everyday. Focusing on operational excellence, I enhanced the rider assignment algorithm and reduced logistics costs by 20% while simultaneously increasing rider utilization by 30%.
• Led the cross-functional effort to integrate the Everyday service into high-visibility areas of the Zomato app, including Quick Delivery, Search, and Homepage.
• Build data infrastructure from the ground up, having designed the baseline Analytics system for Bistro by Blinkit.

Jun 2022 – Jun 2024 · 2 yrs · Gurugram, Haryana, India

• Zomato Everyday - Growth and Expansion
• Created an MIP Solver Model using OR tools to predict cloud kitchen locations for Zomato Everyday, opened 50+ kitchens across India
• Integrated OSRM APIs for s2cell distance & duration; Used app opens, edt, search & supply factor for geographical score computation
• Optimised city-level profitability, making 20+ pods profitable within 6 months, monthly revenue over INR 60 lacs/pod
• Analytics and Modeling
• Conducted A/B testing for product features, set analytics pipelines on Kafka, dashboards for monitoring & analysis
• Deployed an XGBoost MLflow model for KPT with 81% accuracy, set up Grafana dashboard to track performance
• Created T+2 demand forecasting model, reducing out-of-stock issues, saving INR 1.4 lacs/day in yield & order loss
• Developed Packaging model for Zomato Instant to optimise packaging saving INR 4 lacs/month

May 2021 – Jul 2021 · 2 mos · Mumbai, Maharashtra, India

• Project:- Developed an AI model with 3 sub-modules for estimation of banana leaf value based on the quality of wrappers carved out of it
• Created a Leaf extraction model using watershed transform, graph cut for extraction module & achieved precision & recall of 0.92 & 0.91
• Applied Sliding window algorithm for real-time optimum wrapper placement that minimizes used area & maximizes the number of wrappers
• Applied SVM, RF to classify extracted wrappers into quality groups with 83.5% accuracy using Haralick texture, peculiarity, color pigment
• Integrated all the sub-modules into a single architecture and projected 30% reduction in time & labor cost of organic wrapper producers

Sep 2020 – Apr 2021 · 7 mos

• Navigation Path Detection for Autonomous Tractor
• Developing an algorithm for navigation path detection of an autonomous tractor using Image processing and Deep Learning.
• Machine vision, LiDAR sensor, and RTK-GNSS-MARG were combined to create a robust navigation aid system for autonomous navigation.
• Segmentation of image into road, boundary, and vegetation using Open Computer vision in Python.
• Edge Detection for vision-based navigation using K-means Algorithm and Canny edge discrete operator.
• Performed Linear Regression(OLR) and Hough Transform(HOU) to detect the navigation path.

Apr 2020 – May 2020 · 1 mo

• Noisy Channel for Low Resource Grammatical Error Correction
• Formalised GEC in Noisy Channel framework and generated confusion sets from Wikipedia edit history to estimate the channel model
• Combined pre-trained language model (BERT &GPT-2) with the channel model and applied Beam search for optimal output sentence
• Achieved F0.5 score of 40.17 for all error types; observed best results for inserting missing commas, spelling &subj-verb agreement errors

Dec 2019 – Jan 2020 · 1 mo · Bengaluru Area, India

• Developed a grain detection model using Image Processing, Watershed algorithm and extracted the images of each grain using OpenCV
• Performed Image filtering, thresholding, and computed the bounding boxes for contours to analyze the physical parameters of the grains
• Implemented the Sequential models like VGG16, RESNET50, Inception-v3 for grain classification and achieved the accuracy of 71%
• Applied Dropout technique, Batch Normalization on the classification model that resulted in 3.2% growth on overall model's performance
• Successfully combined and deployed the first-grain classification &detection model for the Quality Control Inspection Team of Superzop

Jan 2019 – Feb 2019 · 1 mo

• Analyzed data consumption of more than 78000 network towers and their respective user statistics to predict Congestion in Network
• Applied Maximum Vote Classifier on SVM-RBF kernel, XGBoost, and LightGBM to create robust model followed by feature engineering
• Achieved the MCC score of 0.78 by tuning the hyperparameters of the algorithm using Bayesian optimization and grid search.
• Performed K-fold Cross-Validation with 5 folds to check the robustness and performance of the model which gave the accuracy of 80.46%

Sep 2018 – Oct 2018 · 1 mo

• Employed Arima Time Series Model to the given data of sales of medicines in order to predict the sales pattern of the upcoming 12 months
• Deduced interpretation from the analyzed data to device Optimal Inventory Management strategy to increase profits and sales
• Used Failure Modes and Effects Analysis(FMEA) to look for the probable failures in the model and analyzed them to improve reliability
• Proposed 19 days reorder cycle to reduce the risk of getting out-of-stock and decreasing the holding cost to maximize future profits