Reeti Pandey

APPLIED SCIENTIST II | AMAZON, SEARCH RELEVANCE | JUNE 2025 - PRESENT • Improving the quality of search on amazon.com APPLIED SCIENTIST II | AMAZON, CONSUMER ELECTRONICS TECH | MAR 2022 – JUNE 2025 Seattle(WA) • Generative AI Application : Developed and deployed Generative AI application for internal employees to convert their task demonstration videos into Standard Operating procedure for agentic automation. Used opencv for scene detection and Claude 3.7 for frame analysis and sop generation. Deployed using AWS-CDK using Lambda, Step Functions and API gateway. • Attach Recommendation – Designed a customer facing experiment to test a Graph Neural Network based model to recommend most relevant accessories for a base product. The ranking algorithm increased product type diversity from 2 to 3. • Bundle Ranking – Developed a two- tier product ranking model, finetuned a Mixtral 8*7B LLM to predict conversion of product bundles using historical data of conversion rates and Claude 3 to generate instruction QnA datasets for training the LLM encoder. Achieved a NDCG@3 score of 0.8 and MRR of 0.7. Model was deployed using AWS step functions and EC2. • LLM Generated content for Bundles - Led several A/B experiments from inception to completion by engineering prompt for Claude LLM to generate content for thousands of bundled-product detail pages, in 5 different languages, achieving a projected annualized revenue gain of 50M USD in US , and 30M USD for other marketplaces for Amazon. • LLM Content Evaluation – Evaluated and validated LLM generated content at scale prior to launching customer facing experiments. Designed quality criteria and tooling to quantify the content quality of LLM generated content for 5 languages. Established 80% acceptable content as go-ahead criteria for GenAI applications.

APPLIED SCIENTIST INTERN | FLIPKART RND | SEPT 2021 – FEB 2022 | Seattle(WA) • Decision Assistant Chatbot (Flippy) : Developed various NLP models for subtasks like chit chat conversations, product recommendations, Question answering etc. • Semantic Text Similarity- Trained a Bert-like text similarity model to identify a similar pair of user queries based on domain specific context. This model helps cluster similar queries, identify new query patterns and Augment training data. Model in production.

SENIOR DATA SCIENTIST | ZENON | AUG 2020 – DEC 2020 | Remote(India) • Supplier Risk Score – Collaborated with clients to create analytics-powered Supplier Risk Management framework using Python scripts to channel JnJ supplier evaluation metrics from external agencies and augment them with internal supplier records using text matching algorithms on their names and addresses with a match rate of 25%.

Business-minded Data scientist with 3.5 years of experience and demonstrated ability to deliver valuable insights via data analytics and advanced data-driven methods, and develop Machine Learning Solutions for various Healthcare stakeholders. -> Developed a Computer Vision and Transfer Learning based solution for medical image annotation tasks that automated and sped up the processing of medical claims and saved up to 1.5 FTE a year for UHC. -> Developed automation for the task of Rx pre-authorization process that replaced the rule-based determination process with a decision-tree-based determination process that generated cost savings of 1 FTE a year.

-> Developed an NLP-based solution for Sarcasm detection as a POC for my team. The objective of the POC being showcasing our NLP abilities to our prospective clients. -> Developed a Recommender system for Liveandworkwell.com members to recommend suitable providers to the members using their claims data. -> Developed dashboard for a client delivering valuable insights using their transaction data.