Neha Duggirala

NLP Data Scientist

Shell•  October 2022 - Present•  Remote

Built InvoiceCraft, a Shell E LLM-powered solution that converted unstructured invoice data into structured JSON. Integrated with a chatbot for querying and real-time error correction, allowing users to modify the JSON output of invoices directly and save it, reducing manual processing costs by 68%. Designed a powerful search tool for M&A and CS, including Microsoft rank API to boost ranking and indexing, enhancing information retrieval efficiency. Additionally, established an Azure deployment pipeline for Findr, setting up the entire workflow, and deployment from scratch for the Findr Dashboard in Azure, ensuring a fail-proof system. Independently managed stakeholder relationships for the integration of an Enterprise Resource Planning system, enhancing data accessibility and streamlining processes across departments. Implemented a foundational inner source evaluation project for company-wide search applications at Shell. Developed an evaluation system with metrics such as precision, recall, DCG, and IDCG, creating a framework to benchmark and compare the performance of different search systems on various datasets. Evaluated the effectiveness of RavenPack Text Analytics services, specifically Text extraction, NER, and Event Sentiment, in leveraging the API for internal projects in Shell.

Data Science Analyst

Cardlytics•  July 2020 - October 2022

Developed a Streamlit application in Python to identify and extract 98% from MasterCard brands and other data sources. Created a full automation pipeline that fetches transactions daily using a cron job and processes to identify brands, reducing the time required from 2 days (previously done manually by 5 analysts) to just 30 minutes with minimal manual intervention. Using clustering algorithms (KMeans, Hierarchical, DBSCAN, HDBSCAN) and Char encoding to effectively categorize 40% of complex bank transaction data, improving campaign building.