Publication Date

Spring 2026

Degree Type

Master's Project

Degree Name

Master of Science in Computer Science (MSCS)

Department

Computer Science

First Advisor

William Andreapolous

Second Advisor

Nada Attar

Third Advisor

Sayma Akther

Keywords

large language models, retrieval-augmented generation, long-term stock investment analysis, anomaly detection, fraud-risk signaling, explainable AI

Abstract

A large number of financial information such as annual reports, financial statements, and market data is readily available to retail investors. However, the tools for transforming these content into a useful investment strategy are generally not available to retail investors, or at best are superficial. This project builds an LLM-based Investment Advisor Assistant to bridge this gap by providing three distinct modules : Document-based Question Answering over 10-K filings, Fundamentals-based Investment Scoring, and Market Anomaly and Fraud-Risk detection. It incorporates a Retrieval Augmented Generation (RAG) pipeline for search over the companies’ financial reports, an engine to perform fundamental analysis using profitability, cash flow, and balance sheet metrics, and an anomaly detection system to detect potential manipulation of stock prices using Isolation Forest, DBSCAN algorithm, an auto-regressive baseline and and a transformer autoencoder. The output from these three modules feeds into an advisor layer where it will be integrated as evidence-backed investment recommendation delivered to the end user through a Streamlit-based interface. Overall, this project illustrates how deep understanding of financial documents, quantitative analysis, and LLM-based inferential reasoning can be effectively combined into a single, transparent system that can assist in performing investment research.

Download

Available for download on Saturday, May 22, 2027

Share

COinS