Aug 2022 – Dec 2023 · 1 yr 4 mos · Singapore

• 1. Responsible for improving the Machine Learning infrastructure by introducing centralised feature pipeline and automating the processes of monitoring feature distribution shifts.
• 2. Collaborated with multiple teams to improve the experimentation efficiency of models which included improving our A/B test platform as well the model inference pipeline.
• 3. Lead the development and experimentation of multiple Ads CTR models by introduced semantic keyword search embeddings and cross architectures such as DCN-v2 for the food-search scenario of GrabFood in SG/ID/PH/MY/TH/VN. Overall the clicks improved by 8% and ads-revenue improved by 10% (~5 million usd yearly)
• 4. Collaborated with the Search and Personalisation team of GrabFood to improve the keyword-recall quality when user search for food on the app. The collaboration effort improve the clicks for the ads scenario as well.

Oct 2021 – Aug 2022 · 10 mos

• 1. Responsible for research and study of user-behaviour sequence modelling to predict which item user is likely to click next.
• 2. Responsible for implementing state-of-the art models such as YoutubeDNN and MIND(Multi-Interest Network with Dynamic Routing) and moving the models to the inference pipeline and conduct A/B test in the live environment. The models improved the daily CR by more than 5%
• 3. Responsible for maintaining and improving various machine learning systems components such as visualization system, offline inference pipeline and serving pipeline.

Dec 2020 – Oct 2021 · 10 mos

• 1. Responsible for building learning-to-rank models using multi-task learning for the similar product scenario for Shopee's ecommerce business scenarios and improved the CTR and CR metrics in the online environment by 3%.
• 2. Responsible for modelling and training product text embeddings using discriminative learning with Roberta model. This work improved upon previous tag-matching models for context-based recall and work-embedding based models and was able to significantly improved the CTR for the similar product scenarios.
• 3. Handled various projects related to NLP such as named-entity recognition, machine translation etc.

Jan 2019 – Dec 2020 · 1 yr 11 mos

Jul 2018 – Sep 2018 · 2 mos · Hangzhou City, Zhejiang, China

Aug 2017 – Dec 2017 · 4 mos · Singapore

Feb 2016 – Jul 2017 · 1 yr 5 mos

• Item-Item Collaborative Filtering(Recommendation System):
• Item-Item based collaborative filtering to provide suggestion for lenses, shades and glasses for an ecommerce firm, Lenskart. Indexing was used to cope up with sparsity in dataset and also to increase density of data points. Implementation of an iterative approach to provide scalability by calculating item-item similarity coefficients’ by formalizing a temporal stochastic version of Pearson’s coefficient.

Feb 2016 – Jul 2017 · 1 yr 5 mos

• Android XMPP Notification Sender: As a key team member, designed and implemented an android based notification service using XMPP protocol. (Tools used: Java, Mongo DB, MySQL, Kafka Pub/Sub, Kibana, Elasticsearch)
• Kafka Message Sender: Design and implementing end-to-end rich segmentation-based message sender service for android and iOS devices. (Tools used: Python, Mongo DB, MySQL, Kafka Pub/Sub, Kibana, Elasticsearch, Aerospike)
• Product Feed Notification Engine: Responsible for analysing changes in product prices of various ecommerce clients such as Lenskart, Lamudi etc, and designing a segmentation algorithm that sends rich price drop notification to a set of users which were relevant and chosen via a robust segmentation algorithm.
• Responsible for managing the AWS servers, including implementation of alerts in case of overutilization of resources such as CPU and memory. Additionally, to realize the importance of automatic integration, integrated Jenkins to the codebase.

Jun 2015 – Oct 2015 · 4 mos

• Skills Learnt: Python, Google App Engine, NoSql Database.
• Product Name: HappierHr
• Project Description:
• HappierHr is a global Human Resource Management Software aiming to simplify the recruitment process in Hiring Personnel.
• Tasks involved designing the architecture of the application, designing the structure of the data of various entities like applicants, interviewers, HR personnel and other users. Importantly, how all these entities would interact with each other in the application.
• 1. Applicant Add and Edit Functionality.
• 2. Applicant List VIew, Functionality.
• 3. Interview Scheduling and Rescheduling Functionality.
• 4. Jobs Add and Edit Functionality
• 5. Duplicate Applicants can be added. Designing an Algorithm to group duplicates.
• 6. Merge Duplicate Functionality.(Whether multiple duplicates would be merged into one)
• 7. Decoupling of Applicants. (In cases where the algorithm of decoupling wrongly identifies an applicant to be a duplicate of another applicant).

Aug 2014 – Jan 2016 · 1 yr 5 mos

• Skills learnt: Python, NLTK, Scikit, Text Extraction, Preprocessing of Raw Data, Decision Tree , Conditional Random Fields , Regular Expressions , Concepts in NLP
• Building a Resume Parser as a part of HappierHr, a HR Management software, to help HR parse Candidates Resumes. Resume will be uploaded in Pdf, Doc, Docx and parser parses and returns the following fields :
• Name, Email, Mobile Number, Date of Birth , Address, Educational Institutes, Skills , Location, Orgnization Details, Designtion, Duration etc.
• A short Summary:
• Marking a line from a resume as a section, sub-section or neither is a Classification Problem, we chose the Decision Tree Learning Model which prooved quiet fast.
• For the purpose of extracting Organiztion Names , Designation in the Company, Duration etc. from the resume, we used Conditional Random Fields.
• .

Aug 2014 – Dec 2014 · 4 mos

• Skills learnt: ASP .net, C#, msSql, Google Oauth, Google APIs , Android
• Worked with a client (A leading Bank) on creating a service that uses Google Now API to notify Bank's customers with relevant feeds from their bank account in their Google Now App. The Service included an Android App for users to register and a ASP .net Server which was use to create and post cards into users feed with information provided by the Bank. The android side of the project was designed and implemented by me. Server was implement in ASP .net , with .net 4 framework and and MVC style of architecture and msSql Data bases.Currently working in a project on Google Now .

May 2013 – Jul 2013 · 2 mos · Bangalore

• Technologies Used : C# , ATL Librries , Microsoft Foundation Classes, C#,.NET ,VC++,USB Connection Protocol.
• Printer Simulation Framework(PSF) is an in house product used to provide simulated printer devices for the corresponding hardware device connected over the USB port. In order to achieve this information from the actual hardware should first be extracted and stored in the simulator database.
• Cloning printing devices for simulation to accomplish the task of fetching information from real hardware , processing it in the simulator database .This project involved technologies such as COM,
• Network Communication Major Implementation involved UX Design and Communication with the
• Hardware , Information Processing .

May 2012 – Jul 2012 · 2 mos · Kolkata Area, India

• Implementation of an Employee Management System to aid the employees to access and edit their various information.
• The personal data of the employees were stored on a Mysql database
• JSP,JDBC with JBOSS Server was used as the front end.