Jul 2022 – Mar 2024 · 1 yr 8 mos

• Working on improving Search Relevance and Ranking by training two-tower model on user interaction, product content and contextual data over the platform.
• Developed a virtual chat agent to cater to the post-order customer queries using LLM, gpt-4 and intent detection using llama index. The chat agent is deployed on production serving 95% daily chats and 30% reduction in customer support cost.
• Developed and deployed catalog-level multimodal embeddings using product images and textual information for ~50M active catalogs on platform, which lead to 0.7% improvement in Order By View (O/Vi) values. These are also being utilised in Search models for retrieval and ranking tasks.
• Created and deployed multi-output image classification models to extract visual attributes from product images uploaded on Meesho platform, resulting in ~99.6% reduction in taxonomy time & ~1.59Cr INR savings in avg. listing cost for products
• Developed an incremental and scalable algorithm for grouping ~4 Billion product pairs based on image similarity scores derived from product image embeddings, the resulting group tables are being used to identify duplicate products on Meesho, in Price Reco Tool to suppliers, in catalog compliance and more. This resulted in improvement of NPS and VRS coverage up by 8pp

Jul 2021 – Jul 2022 · 1 yr

• Responsible for researching & building algorithms to understand the Offline, O2O (Online-Offline) customer behaviour utilising their geo-location data
• Assisted partner marketing campaigns (Vissel Kobe, Eneos, Starbucks) by creating target user segments based on users geo attributes and personas
• Expanded the primary point-of-interest dataset by 11.95% by improving the class labels using multi-class classification and hierarchical clustering

Mar 2021 – Jun 2021 · 3 mos

• (Received Pre-Placement Offer based on exceptional performance during the internship)
• Generated location-based & description-based embeddings using Loc2vec & pre-trained transformer, combined them using auto-encoders
• Predicted next visit location of users using a multi-embedding GRU-based RNN model utilising the generated embeddings and users’ trajectories

May 2020 – Jul 2020 · 2 mos · Bengaluru, Karnataka

• (Received a Pre-Placement Offer (PPO) for Data Science Associate Position)
• Developed a Social Listening Pipeline to monitor different social media channels for any direct brand mentions, customer feedback, specific topic discussions
• Pre-processed Twitter dataset by filtering tweets of HCPs using various custom filters, cleaning filtered tweets followed by keyword/phrase extraction
• Employed many semi-supervised and unsupervised topic discovery approaches to identify the topics being discussed in a semi-automated manner
• Generated insights from identified topics using Extractive summarization, Sentiment and Subjectivity Analysis, topic trends with time and geography
• Tech Used: NLP, Spacy, Embeddings (USE, Fasttext, InferSent, Sentence Transformer), Association Graphs, LDA, Corex, Zero-Shot BERT Model, Optics Clustering, Ipywidgets

Jan 2020 – Apr 2020 · 3 mos

• Conducted multi-faceted research for the K-12,higher education & executive education markets in Philippines, Malaysia, Indonesia, Singapore, Thailand, India, Saudi Arabia & China
• Supported in principal expert recruiting, Web scrapping, and market research tasks.

Dec 2019 – Dec 2019 · 0 mo · Gurgaon, India

• Tested Customer Credit Worthiness by analyzing user’s text messages, location and app-usage data collected from Hora mobile application
• Created a Salary messages identification algorithm using a rule-based approach including regular expressions and NER to estimate the user capabilities
• Built geographic models to determine major user locations, potential home, and office. Used ATLAN Grid api to find affluence index of these locations

Jun 2019 – Jul 2019 · 1 mo · Tokyo, Japan

• I went for a 6-week long internship through the University of Tokyo Summer Internship Program. I worked under Prof. Kazuo HIEKATA on Implementation of the Internet of Things using Cloud Computing services like Microsoft Azure and Power BI.
• The following objectives were accomplished:
• ● Proposed a reference architecture for the visualization and analysis of real-time temperature and humidity values which can be followed by non-IT experts for analysis and experimentation.
• ● Used services of Microsoft Azure like Azure IoT Hub and Stream Analytics to connect to Raspberry Pi and analyze real-time data.
• ● Organized a small hands-on IoT workshop for all the lab members in which they worked on individual raspberry pis for connecting to Azure services and experimentation.
• UTSIP Program: UTSIP Kashiwa, initiated by the Graduate School of Frontier Sciences in 2013, provides a hands-on research internship program to approximately 39 students from all over the world in total in the departments of Biology, Chemistry, Environmental Studies, Information Technology, International Studies, Materials Sciences, Advanced Energy, Ocean Science, Physics and Sustainability Science Studies. On the final day of each UTSIP Program, participants give a presentation summarizing their research and findings from the program.

May 2018 – May 2019 · 1 yr

• Pre-processed and merged 3-years of time series Sensor Data of different Shell Assets/Equipment.
• Worked with a faculty of IIT Kharagpur on a year-long consulting project with Shell, India on anomaly detection and 'Remaining Useful Life' (RUL) prediction using Machine Learning techniques.
• Used pattern recognition to recognize the health state of the equipment by studying the similarity of time series sensor signals based on Hidden Markov Models (HMM).
• Predicted Remaining Useful Life (RUL) of the assets using Polynomial Regression and Support Vector Regression (SVR).

Mar 2017 – Dec 2018 · 1 yr 9 mos · Kharagpur Area, India

• Kharagpur Robosoccer Students' Group (KRSSG) is a student-run research group under the guidance of professors of IIT Kharagpur and aims to develop a team of autonomous soccer-playing robots. The group also participates in the prestigious RoboCup Competition held on an international platform under the team name KgpKubs.
• As a mechanical team member, I worked on the design, fabrication, and optimization of the locomotion system, kicking and dribbling mechanisms of the robots.
• Spearheaded the designing, manufacturing, and procurement of different parts worth INR 300,000 for Small Size League Bot, Robocup.

Aug 2016 – Oct 2017 · 1 yr 2 mos · Kharagpur Area, India

• Managed matters related to sponsorships, social media content and media relations of the Formula SAE team of IIT Kharagpur.