ARJUN DEBNATH
Summary
Data science professional with 8 years of experience in designing solution architectures and productionizing ML/DL solutions on cloud enabled systems (GCP/AWS). He has extensive experience in BFSI, E- commerce, Media and Space systems industry. I am also a Kaggle competition expert and take particular interest in implementing research papers to production.
Overview
10
10
years of professional experience
Work History
Senior Data Scientist
Truecaller
04.2022 - Current
- Leading efforts to improve fraud detection models for fintech clients. Created PySpark pipelines to analyze datasets encompassing billions of rows with high efficiency and scalability. Optimised existing pipelines resulting in reducing~25 % of cost. Our models helped improve the fraud detection rate by 8 %
Senior Data Scientist
Prudential Vietnam Assurance Private
Ho-Chi minh, Vietnam
07.2018 - 06.2022
- Spearheading the data science team on projects related to agency and partnership distributions for a global life insurance company
- Agency Cross-sell: Successfully implemented a solution which targets potential existing customer to cross sell the right insurance product by allocating the right agent for Prudential Vietnam. This increased the overall Annual premium equivalent (APE) from 1.5 % to 7% across 3 quarters
- Real time ESG news classification: Developed a lightweight XGboost model using TF-IDF embeddings which classifies a news content to either ESG relevant or not and achieved an accuracy of >85 %. FastAPI on heroku is used to host the App
Principal Data Scientist
Kawa Space
04.2020 - 04.2022
- Leading the data products team at KawaSpace to build a scalable geo-intelligence platform for Fintech customers
- Create a scalable land boundary detection model on satellite images using Tensorflow. Responsible for creating end to end data pipelines which consumes satellite images , preprocess and performs online predictions on GCP
- Innovated a data standardisation algorithm for storing and querying satellite imagery thus enabling 5x improvement in latency
- Developed a customer segmentation model for a Fintech client using alternate data which enhanced loan default identification by 10 %
Machine learning engineer
Quantiphi Analytics
06.2017 - 07.2018
- Content features extraction: Developed an NLP framework to extract from a Corpus of Text using for a Leading Media client which helped them understand various KPI’s impacting the viewership of a TV channel thus improving theirTRP by 5%
- Normalisation of Product descriptions: create a text normalisation algorithm using windowed fuzzy logic and Parsers which standardises apparel features from multiple retailers for a leading Online shopping client resulting in better recommendations to its customers
- Developed a Text-Image Hybrid classifier model which maps an apparel to its category which increased the accuracy of apparel class detection by 15 %
Business Analyst
EXL Service
02.2016 - 05.2017
- Developed a risk model for a Leading Banking client using Logistic Regression and PCA, which predicts and declines a possible Credit Card Fraud transaction during real time
- Developed, tested and implemented a time series forecasting model which would predict the monthly and the daily losses for a leading bank, resulting in 20% increase in predicted accuracy
Education
Bachelor of Engineering - undefined
Jadavpur University
01.2015
Skills
- Python
- Pytorch / TensorFlow
- NLP
- PySpark
- Haystack / LangChain
- Faiss / Pinecone
- Computer Vision
Custom
- Microsoft AI hackathon - Given a user query and candidate passages corresponding to each, the task is to mark the most relevant passage which contains the answer to the user query.An LSTM model was created to get top k accuracy of 70 %. Ranked top 10 % in the competition - https://github.com/Peaceout21/MSAI-hackathon/blob/master/lstm-encodermodel.py
- Kaggle | IEEE Computational Intelligence Society - Fraud Detection - Created a stacked ML model to identify fraudulent transactions achieving a mean AUC score 0.76 - Ranked top 10 % on leaderboard submission
- Kaggle | Los Alamos National Laboratory - Earthquake prediction - Created a sequential model using GRUs to forecast the time of earthquake using time series seismic signals with a MAE of 1.42 seconds - Ranked top 10 % on leaderboard submission
Custom
- Generative AI and LLMs 05/2023 - Present Remote / USA - Built a scalable LLM application for a VC investment firm which gathers unstructured data about a target company's company's Product offering, clientele, financials etc. This has helped improve TAT by 15 % Tech stack - Python, Pinecone, Langchain , GCP
- Developed a custom WakeWord detection for a VoiceTech firm and transitioned them to open source thus saving ~ 90 % of cost with comparable accuracy
Timeline
Senior Data Scientist
Truecaller
04.2022 - Current
Principal Data Scientist
Kawa Space
04.2020 - 04.2022
Senior Data Scientist
Prudential Vietnam Assurance Private
07.2018 - 06.2022
Machine learning engineer
Quantiphi Analytics
06.2017 - 07.2018
Business Analyst
EXL Service
02.2016 - 05.2017
Bachelor of Engineering - undefined
Jadavpur University