Vesselin Grantcharov

Led the implementation of a Retrieval Augmented Generation (RAG) application with a chatbot interface, enabling clients to efficiently gather information from Fitch's extensive corporate research corpus. Focused on retrieval/search performance tuning through LLM-based query transformations and oversaw the development of model evaluations, data chunking, and ingestion.

Established data storage and a query engine using Deltalake/Spark, and set up a scalable architecture for researchers to manage compute instances for econometric analyses. Managed data engineering and provided support for data scientists on the Higher Education Finance project, leading to published research.

Designed a Learning to Rank model to enhance search functionality by incorporating user context, such as view history, beyond pure query-document similarity. Tested and implemented a predictive caching model to reduce end-user perceived latency by pre-loading anticipated data.

Contributed to a quality optimization initiative by developing an LSTM-based model to predict plywood sheet 'grade' after the drying process. This model utilized time series data of dryer temperatures and assembly line velocity.

Served as a Teaching Assistant for Linear Algebra, providing support and guidance to students. Assisted with problem-solving and concept clarification.

Currently pursuing a PhD in Nanoscience and Microsystems Engineering. This program focuses on advanced research in nanoscale and microscale systems.

Activities and societies: Undergraduate research with Ernest Croot in additive and algebraic combinatorics, focusing on the Cap Set problem. Developed software to represent, validate, and construct cap sets using graph search algorithms.

Completed a Bachelor of Science in Discrete Mathematics with a minor in Computer Science, specializing in Intelligence. The curriculum covered foundational concepts in mathematics and computer science.