Mar 2023 – Jun 2024 · 1 yr 3 mos · Hyderabad, Telangana, India

Jun 2021 – Mar 2023 · 1 yr 9 mos · Bengaluru, Karnataka, India

• Amazon Physical Stores Tech Team : worked as a lead software developer and subject matter expert for Scanogram applications, to reduce time taken to find products(like bread, egg etc) by pickers/customers and capture the location of products in the physical stores
• Rack/Bay Numbering Project : helped in improving Units Picked Per hour by 2.5+ , which had 12.5M $ impact in labour saving across 500 stores. I created an algorithm from scratch to automatically assign human readable numbers to each rack in store. Independently worked on HLD, LLD, code for async workflow using S3, SQS, DynamoDB (AWS) and deployed in production.
• Scan Session BI IntegrationProject: exposed the session data to clients and business team for reporting, dashboarding purposes. Gathered use-cases from stakeholder, created multiple data schema option to cover use-cases. Delivered the project.
• In-Store AR Navigation Application: Developed a pilot to find the shortest path to a product and get the directions using AR in IOS devices in less than 2 weeks. This pilot was also added to the product roadmap.
• Worked on creating infrastructure using code, scaling/de-scaling of services, security vulnerability like log4j.
• Mentored one intern and one new hire and have taken 39+ interviews for amazon.
• Transportation Financial System Team: It helps in calculating the shipping cost of a single parcel.
• Delivered enrichment-based cost allocation strategy for GAP service, to allocate cost of shipping using single product shipment strategy instead of just weight. It was a spark job-based service written in java to join and process terabytes of data every month.
• Scaled the services during q4 peak 21' in amazon. Added and improved in total 40 plus alarms and ticket for SQSs and other resources to make system ready for peak traffic.

Sep 2020 – Jun 2021 · 9 mos

• Worked as part of the PayFlow Payment Gateway Solution Teams, which is part of the Unbranded Payments solution for PayPal. PayFlow apps provides Onboarding, Reporting (recurring), Billing (recurring), Configuration tools and different payment methods to the merchants along with custom processor options. We have in total 13 applications some of which are UI based, some API based and some are the base applications to support other applications.
• I worked to roll out a security enhancement to update the SHA1 hashing algorithm to SHA2 hashing algorithm across all the application. It involved coding and testing all the possible scenario's, as missing a scenario could have raised a P0/P1. Also worked on improvements, enhancements and debugging bugs of P2, P3, P4 priority on 7-8 apps on a regular basis.

Jun 2019 – Dec 2019 · 6 mos · Koramangala,Bangalore,India

• Automatic Number Plate Recognition System: Developed and deployed the Automatic Number Plate Recognition by creating a custom two-part deep learning model consisting of a license plate detector and OCR module, using YoLo architecture (DARKNET). Achieved 4-wheeler accuracy –
• 96% and 2-wheeler accuracy – 56%, outperformed by 2-3% other commercially available software.
• Human Compatible Machines: POC to develop a generic framework, with capabilities like - emotion detection, empathy, trust. RASA chatbot was developed and deployed on slack, to generate a pseudo empathic response based on detected human emotions. Also, fine-tuned the Tacotron (Text-ToSpeech) model to produce voice modulated speech with the added emotional tone of amusement, disgust, anger, etc.

Jun 2018 – Jul 2018 · 1 mo · Jaipur Area, India

• Worked on a live project named, OTR-One Touch Retail for Mercedes-Benz Germany. Pandas and Numpy libraries were used to automate the process of insurance claims filed against the service contract agreement after car crashes.
• This process was earlier done by an auditing team hired by Damler, based on whether company has defined a claim in the master file and if not, whether the claim is being given to any user previously using the claim's history.Now we had to streamline the process by grouping together all the relevant parts which can be replaced due to a particular car crash.
• We achieved the automation by converting all the unstructured data into the structured data, cleaning it and then applying concepts of random forest to group together the related parts.