Jan 2023 – May 2023 · 4 mos · Bengaluru, Karnataka, India · On-site

• Built loan approval models for various loan partner brands.
• Used financial features built based on CRIF and Experian reports of customers.
• Built a chatbot for FAQs using chatgpt and langchain.

Sep 2019 – Apr 2022 · 2 yrs 7 mos · Bengaluru, Karnataka, India

• Identification of problem statements where data science can help.
• Explored network effects and demonstrated densely connected Okcredit’s user network clusters.
• Lending: Explored lending experiments and initiated efforts with ola which helped us in having better understanding of the lending space in the country. We ran a cohort based experiment to gauge the need for credit also. Gst tin number extraction from printed bill images to find potential heavy users.
• Voice based action: Built voice based transaction addition MVP.
• Growth: Worked on running cohort based growth ad campaigns. Learned about the impact of cohort size, creatives, organic growth, event based targeting using super user cohort.
• Hiring: Interviewed candidates for the data science and analytics team.
• Building teams: Helped in building data science and analytics team and setting up processes and workflows for efficient delivery of solutions to prioritized problem statements, like roadmap building, interacting with stakeholders from all departments, setting up dashboard data science model’s statistics on superset etc.
• Business category based personalisation: Helped in building category prediction model, deployment, setting up daily airflow jobs.
• Payments fraud detection: with Sahil, Mohsin, helped in building payments fraud detection model, deployment and daily analysis reports. Used lightgbm based model and utilized shap library to identify high impact variables.

May 2017 – Aug 2019 · 2 yrs 3 mos · Bengaluru, Karnataka, India

• Screen scraping and PII masking: Helped in extracting data from tables with miscellaneous structures and building column identification using deep learning based models. Also, worked on identifying strategies to identify PII in data and masking it.
• Wealth Data Enrichment : Worked on CNN based text embedding features and siamese based contrastive learning for Security Style identification, Security normalization, Account type identification.
• Transaction Data Enrichment : worked on creating a mapping for possible variations of the words in transaction data using text embedding.

Jul 2015 – Mar 2017 · 1 yr 8 mos · India

• Identity Based Fraud Detection (Sandeh): worked on prediction of return probability of an order based on the user identity features like username, email, mobile, address. The project also includes the functionality to check the well formedness of an address and based on the output and suggest the corrective measures.
• Prediction of Out of Delivery Area (ODA): worked to build a model based on supervised learning to predict whether an incoming order is ODA (Out of delivery Area) or not. Used tf-idf scores as address features and Naïve Bayes as predictive model.
• Similarity and Spike prediction: worked on a similarity project which focused on preventing bulk orders by customers using fake identities but having textual similarities in email and addresses. Clusters are renewed on a daily basis and clustering happens for a segment of subcategory and pincode. Spike project uses Poisson's distribution to indicate spikes in the number of orders placed in a segment.
• RTO prediction: Helped in understanding various fraud scenarios such as bulk buying, seller self buying, courier fraud and inventory blocking and extracting important features to identify them. Building models using features extracted to distinguish between a fraudulent order (RTO) and genuine order based on fraud likelihood scores from them.

Jul 2014 – Apr 2015 · 9 mos

May 2013 – Jul 2013 · 2 mos · Bangalore

• Smarter Commerce:
• worked with Atul, Udaya, Karthik, Mahesh, helped in building NLP Spoken-Web Based B2B Messaging
• Created a voice site using Spoken Web technology which enabled SMBs (Small- Medium Business) to carry out transactions in regional languages.
• Developed NLP based algorithm capable of translating business messages received from customers to standard business documents with the help of language specific dictionary (e.g. Hindi) and target business language dictionary and vice versa.
• Added registration and authentication features which enables the users to securely access the various functionality supported by the proposed solution.
• Added a catalog creation feature which lets registered users to register their produce along with quantity just by calling this B2B agent.
• Integrated with IBM WebSphere Commerce which enables the registered users to manage their account online also and implemented auction over voice facility.
• Implemented aggregation and splitting using which similar produce from multiple farmers are aggregated under a single produce name in the catalog.