Hi, I am Arnav. Nice to meet you.
I am a 26 year old Data Scientist and Guitarist, currently residing in London, originally from India. As an aspiring dog owner with a passion for data, I find myself at the intersection of analytics and creativity. You've likely stumbled upon my page out of curiosity to learn more about who I am. I invite you to scroll down and gain insights into my professional journey.
EXPERIENCE
DATA SCIENTIST
FINTRU
- Led the development, fine-tuning and deployment of BERT based Question Answering models to automate the extraction of KYC requirements from financial documents, saving 2000 staff hours annually
- Conducted the experimentation and A/B Testing of BM25, SPR and ColBERT retrievers, resulting in over 50 % improvement in inference speed
- Optimized training workflows by integrating PyTorch Dataloaders, implementing experiment tracking with MLFlow, and conducting grid search to fine-tune hyper-parameters, resulting in 15% increase in model F1 score
- Collaborated with Data Engineering and Annotation teams on establishing data gathering and storage processes for model training in Azure Blob Storage
- Investigated use cases and fine-tuning requirements of Large Language Models like Llama 2 and Falcon for Reader-Retriever QA systems and evaluated their performance against in-house models
- Onboarded 3 data scientists to the team and mentored a summer intern on Named Entity Recognition Project
NLP DATA SCIENCE INTERN
FINTRU
- Developed 4 classification models for Text Categorization Task using Naive Bayes, Decision Trees, Random Forest, and Support Vector Machines, addressing data imbalance and introducing rule-based pre-processing framework.
- Leveraged Transformer architecture and BERT models, achieving categorical accuracy more than 97%
- Performed the 5x2 Cross Validation Statistical Test to compare differences in model performance.
DATA SCIENCE INTERN
ERNST & YOUNG INDIA
- Coordinated with the Risk Advisory Team and performed data extraction from SAP/SQL Databases using Denodo.
- Reduced query run-time by 30% by introducing scheduled data caching methods on local systems
- Created 3 real-time Dashboards for Client using above data on Microsoft Power BI and authored the documentation.
EDUCATION
London School of Economics
M.Sc. Operations Research & Analytics
Course Modules - Algorithms & Computation, Data & Statistical Analysis, Data Mining, Deep Learning, Modelling & Simulations
Sep 2021 - Sep 2022National Institute of Technology, Kurukshetra
B.Tech - Production and Industrial Engineering
Course Modules - Operations Research, Production Technology, Machine Design, Fluid Mechanics, Supply Chain Management
Jul 2017 - Jul 2021SKILLS
Python
Azure
JavaScript
DataBricks
Microsoft Office
Docker
KUbernet
Streamlet