🎓 Slide 1 Introduction:
Introduction:
“Good morning everyone. I’m Deepika K, an MBA Finance student. Today,
I’ll be presenting my final project on how Artificial Intelligence, especially
through Large Language Models (LLMs), is transforming financial data
analysis.
This project is titled ‘Talk with Data’, as it introduces a conversational AI
assistant that helps users interact with financial data just like chatting
with a human expert.”
🔍 Slide 2 Objective and Method:
Objective:
“The objective of this project is to develop an AI-powered chatbot that
simplifies complex financial data and makes it more accessible.
Users can ask financial questions in plain language—such as ‘What is the
revenue growth of TCS in 2024?’—and get immediate, meaningful
responses.
This helps non-technical users like CEOs, managers, and analysts make
informed decisions without deep financial expertise.”
Method:
“I followed an applied research methodology using a prototype-based
development approach.
The prototype uses LLMs like GPT-3.5 and Mistral, integrated with SQL
databases.
We used Yahoo Finance data of TCS for the years 2022–2025.
Python libraries like yFinance, Pandas, and SQLAlchemy were used for
data extraction, processing, and storage.
The chatbot interface was built using Streamlit, and LangChain was used
to connect the LLM with SQL queries.”
⸻
🧭 Slide 3 Problem Description:
Presentation Notes:
“Financial data is often presented in spreadsheets, reports, and
statements that are difficult for non-experts to understand.
Currently, most tools provide static reports. There is little interactivity, and
even basic financial insights require technical skills.
This project introduces a solution using AI that allows real-time
conversations with organizational data, making insights dynamic and user-
friendly.”
📚 Slide 4 – Review of Literature(1/2)
Presentation Notes:
“I reviewed tools like IBM Watson, BloombergGPT, and ChatGPT.
While they offer some financial insights, most of them have three major
issues:
1.They require data to be sent to external servers—raising security
concerns.
2.They often produce generic responses not specific to an
organization.
3.They lack true interactivity and cannot deeply analyze context-
specific financial data.
📚 Slide 5 – Review of Literature (2/2)
This project addresses all three issues with a custom AI assistant tailored
for organizational use.”
⸻
Slide 6 – Research Methodology
In this project, I followed an applied research methodology. The main aim
was to create a practical, working prototype—not just to study AI in
theory, but to solve a real-world problem in finance.
🔁These are the Steps I followed:
1. Data Collection
I used yFinance to collect financial data of TCS from 2022 to 2025—
including revenue, profit, and expenses.
2. Data Cleaning & Storage
With Pandas, I cleaned the data and stored it in an SQLite database using
SQLAlchemy.
3. Chatbot Development
I created a simple chat interface using Streamlit, where users can type
questions like “Compare revenue from 2022 to 2024.”
4. AI Model Integration
I integrated two models:
• GPT-3.5 – more accurate and detailed
• Mistral – open-source, runs locally, more secure
Using LangChain, the chatbot connects the AI to the SQL database and
gives smart, readable answers.
5. Testing & Refinement
I tested the chatbot with real queries and compared performance between
the two models—based on speed, accuracy, reasoning, and clarity.
📊 Slide 7 – Survey Findings
Presentation Notes:
“We surveyed 103 participants from banking, tech, and finance industries.
Key findings:
• 68% were aware of AI in finance.
• 62% trust AI when supervised by humans.
• 30% work directly in finance, making their input valuable.
This confirms that there is openness toward AI adoption, but with a clear
need for explainability and human oversight.”
🧪 Slide 8 – User Testing Results
Presentation Notes:
“We tested the performance of two LLMs: GPT-3.5 and Mistral.
We measured:
• Response time
• Accuracy
• Reasoning quality
• Clarity of explanation
GPT-3.5 performed better in complex reasoning.
Mistral had faster local response times and is cost-effective for private
deployments.
Both models gave useful and understandable financial answers.”
⸻
🧱 Slide 9 – Application Architecture
Presentation Notes:
“This is the system architecture:
• A simple UI (Streamlit)
• Data fetched from Yahoo Finance using yFinance
• Stored in SQLite via SQLAlchemy
• AI models (GPT-3.5 and Mistral) connected using LangChain
agents
• LangChain helps the AI reason through SQL data and return
human-readable responses
This end-to-end solution is efficient and scalable.”
💬 Slide 10 – Chat UI & Sample
Presentation Notes:
“Here is a screenshot of the chat interface.
Users can type queries like ‘Show the net profit of TCS in 2023’ or
‘Compare revenue trends from 2022 to 2024’.
The assistant returns the answer in simple terms, often with reasoning,
which makes it easier to understand trends.”
‘
✅ Slide 11 – Findings, Conclusion & Suggestions
Presentation Notes:
Findings:
• Both LLMs give accurate and insightful answers.
• GPT-3.5 offers deep analysis.
• Mistral is affordable and private.
Conclusion:
• Conversational AI makes financial analysis easier and faster.
• It reduces dependence on financial experts for everyday
decisions.
Suggestions:
• Add voice input for hands-free access.
• Use graphs and dashboards for better visuals.
• Add multilingual support for wider adoption.
• Integrate with more data sources like financial APIs or ERP
tools.”
