May 2021 – Jul 2024 · 3 yrs 2 mos · Bengaluru, Karnataka, India

• At Prudential, I am responsible to create Data Science and Machine Learning solutions for the Pulse App launched by. The app is currently active in 13 countries (SEA and Africa).
• Prominent projects include:
• 1. Developed Question Answering platform for customer care agents using OpenAI LLM and RAG frame which will help the agent to find the answers to the customer queries regarding policies and the intricate details faster.
• 2. Developed search and content recommendation capabilities in the PruAmanah platform. Solving cold start problems with Multi Arm Bandit.
• 3. Developed content moderation for text, images and videos posted in the pulse app and communities.
• 4. Built a virtual assistant for exercise pose matching and pose correction suggestions. Estimating the body measurements like hip , waist, hip to waist ratio from few images of a person.
• 5. Condition/disease classification from the tongue image.
• 6. Built field extraction service for various identity cards for 13 SEA and African countries.

Sep 2019 – May 2021 · 1 yr 8 mos · Noida, Uttar Pradesh, India

• At Tokopedia, an Indonesian E-commerce giant, I was responsible for creation of solutions for the Search and Ads teams using Data Science, AI/ ML and Deep Learning approaches to enhance the users' search experience by showing them relevant products and ads on one hand and enhancing the ads reachability of the sellers to increase the revenue for both sellers and the company.
• The challenging part (also the fun part) is most of the models are needed to be created from scratch due to the nature of industry and the language (Bahasa, Indonesian).
• I have leveraged several word embedding generation methods from the scratch like FastText, BERT, ELMo and Transformers. I have also worked heavily on Semantic Nearest Neighbour search. Explored highly efficient storages with semantic embedding search capabilities like ANNOY, FAISS and SCANN.
• My projects include-
• 1. Category predictions for user search queries. The major challenges are the typos, language (mixed Indonesian and English) and a highly imbalance category distribution.
• 2. Finding the relevant keywords for the seller to bid for his advertised products to enhance the reachability.
• 3. Enhancing the Ads fill rate by proposing alternate query to the user's search query to enhance the number of relevant ads clicks and impressions. Also creation of CNN- Siamese based validator to choose the best alternate query.
• 4. Deep Semantic Match model creation for matching search queries to Ads products semantically, where there are no exact word matches between the search query and products.
• 5. Learning to Rank - This project aims at creation of implicit ranking of the returned matching products for a user query according to previous buyer behaviours.

May 2019 – Sep 2019 · 4 mos · Noida Area, India

• 1. The primary projects I worked on were Claims Survey Comments Classification and Customer Pain-point Prioritization using Text Classification, Text Similarity, Natural Language Processing and Deep Learning approaches. This project helped in creation of an automatic pipeline using which claims team know the problem areas and improve them according to the priority.
• My major contributions-
• a. Bringing the Deep Learning framework into the picture which improved the current system by 17%.
• b. Designing a text similarity solution for the find the relevant problem area within the predicted class.
• 2. Another project was based on Detection of Opioid abuse by the patients and clustering the different types of abusers.

Jul 2017 – May 2019 · 1 yr 10 mos · Noida, Uttar Pradesh, India

• 1. Created Process Intelligence suite which involves Predictive and Prescriptive Analytics with analysis of Historical Process Data, estimation of Turn Around Time for request tokens using case variables and machine learning, prescribing optimum resource allocation for cost reduction using multiple simulations and optimization methods.
• 2. Presented our paper 'An end to end approach for Network Flows: Monte Carlo Simulation Methodology' at IISA 2017 (https://www.intindstat.org/iisaconference2017/), an International Conference on Statistics, under the guidance of Dr. Sayaji Hande(www.hande.in).
• 3. Worked on "Text of Interest" identification, validation scoring of OCR extracted invoice fields and document classification using machine learning techniques.

Jun 2015 – Jun 2017 · 2 yrs · Noida, Uttar Pradesh, India

• 1. As a part of the 4 member team, I was involved in the design and development of an Analytics Framework NEAF for the organization which is used to create Event Based Data Flow Topology for data extraction, manipulation, model creation and execution (i.e. data extraction to model evaluation). The framework has in-memory ETL and Analytics(Regression, Classification, Sentiment Analysis, etc.) capabilities.
• 2. Predictive Modelling using Statistical Techniques (Regression, Classification, Clustering etc.) and Machine Learning Algorithms (Linear Regression, Random forest, Naive Bayes, Decision Tress etc.) on the Process Flow data.
• 3. Designed and developed a robust Simulation Engine for Business Process Flow suite from ground up using various statistical methodologies like Queuing Theory, Markov's Chain, Graph Theory and Monte Carlo Simulation methodology. The Simulator gives very intricate details on the health of the process flow, the resource allocation on various nodes, etc. The engine has the provision for What-if analysis on designed scenarios and historical events. This project was guided by Dr. Sayaji Hande.
• 4. EDA and Predictions on Loan Origination data for detection of credit default .