Aashish Sharma

• • Rag at Scale : Led the development of a UI-based application to collect feedback from Subject Matter Experts on responses generated by a Retrieval Augmented Generation (RAG) model for financial queries related to the London Stock Exchange. Utilized Python and Streamlit to build the frontend interface and Azure PostgreSQL to structure and store expert feedback. Deployed the application into production using Azure in collaboration with the DevOps team.

• • NL2API : Leading the model improvements of Abacus AI's RAG model (based on ChatGPT) on Lipper API, a comprehensive database of fund-level financial data. Developed an evaluation pipeline in Python to benchmark

model performance on a curated Gold Dataset. Currently enhancing the model through prompt engineering techniques and refining document retrievers to boost retrieval relevance and response accuracy.

• • House Plan Pricing Model : Developing a time-series forecasting model to predict the base plan price for individual houses : Utilized boosting algorithms including XGBoost and LightGBM - reduced MAPE error by 57% : Employed Snowflake to efficiently manage and store a dataset comprising 20 years of US house plans data : Used SQL for data processing and ETL pipelines

• • Plan CMA Model : Spearheaded the development of an unsupervised learning method to identify comparable properties for a subject plan based on location and house properties : Implemented advanced clustering algorithms to efficiently group similar properties, enhancing accuracy and scalability : Deployed into production using AWS and Docker

• • Collaborating closely with cross-functional teams including MLEs and product managers to ensure the accuracy and relevance of model outputs : Analysed real estate datasets, utilising statistical methods to present actionable insights to non-technical stakeholders using python libraries Matplotlib and Plotly

I was working with DataKirk as a Data Science Tutor, giving lessons on implementing Data Science solutions in Python.