Jan 2020 – Aug 2022 · 2 yrs 7 mos · Bengaluru, Karnataka, India

• Led team of 2-4 data scientists working on applied machine learning and operations research problems. Was responsible for conceptualization, implementation and deployment of all data driven solutions. Selected work:
• Serviceability
• Designed and developed a serviceability model that predicts the optimal number of orders to be accepted from a client using demand forecasting and supply calculations.
• Allocation
• Designed and developed an order allocation model using a vehicle routing solver that assigns hyperlocal orders to riders to optimise for efficiency while handling multiple constraints and penalties.
• MLOps
• Deployed an MLOps platform for all ML needs: experimentation, tracking, orchestration and model deployment.
• Runsheet Creation
• Developed an automatic runsheet creation model based on geocoding and past data.
• Supervised development of several ML models, including mask/sanitiser detection in selfies, anomaly detection in time-series data, order cancellation prediction, and intent recognition from call transcripts.

Apr 2018 – Jan 2020 · 1 yr 9 mos · Bengaluru, Karnataka, India

• Geocoding
• Developed a geocoding model based on named entity extraction and matching to historical data.
• Developed an NER model to extract named entities from unstructured addresses.
• Risk Profiling
• Developed a risk profiling model to identify high-risk riders using survival analysis.
• Simulation
• Designed and developed a discrete event based simulation environment to model the entire hyperlocal ecosystem, including serviceability requests, orders and rider behaviour.
• Vehicle Routing
• Implemented a vehicle routing algorithm to optimise linehaul routes and assign orders to the most optimal route.
• Demand-Supply Matching
• Designed and implemented a novel demand-supply matching algorithm to selectively accept hyperlocal orders that lead to higher efficiencies.

Mar 2016 – Mar 2017 · 1 yr · Bengaluru, Karnataka, India · On-site

• Face Recognition
• Implementation and comparison of various face-recognition techniques (including deep-learning) for intelligent face-based search in image galleries.
• Samsung Keyboard Project
• Design and implementation of an RNN-based character-level language model based on study of the existing word-level language model.

Jun 2014 – Feb 2016 · 1 yr 8 mos · Bengaluru, Karnataka, India · On-site

• Quantifiable Fitness Tracking using Wearable Devices
• Development of an app using sensor data from wearable devices to track the user's physical activity, estimate calorie expenditure and calculate a heart-rate based fitness measure called Endurance.
• Design, training, validation and testing of a neural-network based classifier for activity recognition.
• The research paper was presented at EMBC '15, Milan.
• Remote Health Monitoring System for Detecting Cardiac Disorders
• Analysis and tuning of the SVM classifier to optimise trade-off between specificity and sensitivity
• Published in Systems Biology, IET, vol. 9, no. 6.
• Fast Non-Blind Image Deblurring with Sparse Priors
• Comparative study of non-blind deconvolution techniques given a blurry-noisy image pair.
• The paper was presented at CVIP'16, Roorkee.

May 2013 – Jul 2013 · 2 mos · Bengaluru, Karnataka, India · On-site

• Keyword Extraction Using Ontologies
• Development and testing of unsupervised algorithm combining semantics extracted from ontologies such as WordNet and graph ranking techniques to select keywords from unstructured text documents.