Retrieval Augmented Generation RAG-related ideas for Master Thesis

Hi Everyone,
I am studying RAG (Retrieval Augmented Generation) and looking for some ideas for my Master thesis project in Data Science. My background is Computer Science bachelor degree and I really excite to jump into Banking & Finance field as well.
Could anyone suggest me some ideas for my thesis which combines RAG and Banking/Finance together?
Really appreciate!!! Thanks!

Six Thesis Ideas

1. Intelligent Financial Advisor RAG System

Build a RAG system that retrieves and synthesizes personalized financial advice from regulatory documents, market reports, and investment guidelines.

Key Components: Multi-modal retrieval (text + tables), regulatory compliance verification, risk assessment integration

Data Sources: SEC filings, Fed reports, bank policies, investment prospectuses

Research angle: Compare retrieval strategies (dense vs sparse vs hybrid) - which one actually works better for financial accuracy? Could make a good comparative study.


2. Credit Risk Assessment with RAG

Enhance credit scoring by retrieving relevant historical cases, economic indicators, and regulatory changes for context-aware risk assessments.

Key Components: Temporal-aware retrieval, fairness metrics, explainability framework

Data Sources: Credit bureau data, economic indicators, default cases, policy changes

Research angle: Bias mitigation in RAG retrieval for fair lending compliance - this is important and timely given all the scrutiny around algorithmic fairness in finance.


3. Fraud Detection RAG Framework

Build a RAG system that retrieves similar fraud patterns, regulatory alerts, and transaction contexts to improve real-time detection accuracy.

Key Components: Real-time retrieval, anomaly pattern matching, cross-institution knowledge sharing

Data Sources: Transaction logs, fraud case databases, regulatory warnings, news articles

Research angle: Privacy-preserving RAG using federated learning across banks - challenging but very relevant for industry collaboration.


4. Market Intelligence RAG Platform

Create a RAG system for financial analysts that retrieves and synthesizes insights from earnings calls, news, analyst reports, and market data.

Key Components: Multi-source retrieval, sentiment analysis, temporal grounding, fact verification

Data Sources: Earnings transcripts, Bloomberg/Reuters feeds, analyst reports, social media

Research angle: Does retrieval quality actually impact investment decisions? Would need some creativity to measure this.


5. Regulatory Compliance Q&A System

Build a RAG system that helps banks navigate complex regulations (Basel III, GDPR, AML) by retrieving relevant clauses and providing compliance guidance.

Key Components: Legal document parsing, multi-jurisdictional retrieval, citation tracking

Data Sources: Basel accords, local banking regulations, case law, compliance handbooks

Research angle: Hierarchical retrieval for nested regulatory frameworks - regulations reference other regulations, so this could be interesting structurally.


6. Personalized Banking Assistant

Build a customer-facing RAG chatbot that retrieves user transaction history, product info, and financial education content for personalized support.

Key Components: User-specific retrieval, privacy-preserving embeddings, multi-turn dialogue

Data Sources: Transaction data, product catalogs, FAQ databases, financial literacy content

Research angle: Measure RAG hallucination risks in high-stakes financial contexts - critical for deployment but not always studied carefully.


A few things to nail down early:

Research Questions:

  • How much does RAG actually improve over baseline LLMs in your domain?
  • Which retrieval strategy works best for financial data?
  • How do you measure trustworthiness and reduce hallucinations?
  • What are the latency/cost trade-offs?

Evaluation Metrics:

  • Retrieval precision/recall
  • Answer accuracy & factuality
  • Response time & cost
  • Domain expert evaluation (try to get some bankers/analysts to review)
  • Regulatory compliance rate (if applicable)

Datasets:

  • SEC EDGAR filings (publicly available, good for finance)
  • Financial news from Reuters/Bloomberg (might need institutional access)
  • Banking product documentation
  • Regulatory text corpora
  • Financial Q&A benchmarks like FinQA or TAT-QA (good for baseline comparisons)

Master’s Thesis Ideas Collection.pdf (92.7 KB)

Pick something where you can actually get the data and has a clear success metric.

Good luck!