VisiCalc introduced for the Apple II in 1979
Lotus-123 introduced for the IBM PC in 1983
Microsoft Excel introduced for Windows in 1987
*Except that they sometimes think it’s OK to use synthetic data without telling us, instead of the data they are instructed to use.
We’ve been developing capabilities powered by Claude since 2023 within AIA Labs.
Claude powered the first versions of our Investment Analyst Assistant, which streamlined our analysts’ workflow by generating Python code, creating data visualizations, and iterating through complex financial analysis tasks with the precision of a junior analyst.
As AI models begin to handle underwriting, compliance, and asset allocation, the traditional architecture of financial work is undergoing a fundamental shift.
As job descriptions evolve, so does the definition of financial talent. Excel is no longer a differentiator. Python is fast becoming the new Excel.
But technical skills alone will not cut it. The most in demand profiles today are those that speak both AI and finance, and can move between legal, operational, and data contexts without losing the plot.
In class, iteratively develop a prompt to get the best performance from a chatbot on a valuation case
I want to set expectations before trying them. Mine may fail, but they won’t be the worst ever.
Read the uploaded case. Ignore the valuation method described in the case. Instead, generate pro forma statements and perform a two-stage DCF analysis.
Document your assumptions and your reasons for them. Generate an Excel workbook and a Word doc. Format the Excel worksheets nicely.
Using the latest version of yfinance, get closing prices for SPY, IEF, and GLD at a monthly frequency since 1990.
Compute returns as percent changes. Compute the sample means and covariance matrix. Compute the mean-variance frontier.
Assuming a monthly risk-free rate of 0.04/12, calculate the tangency portfolio. Plot the mean-variance frontier, SPY, IEF, and GLD, the tangency portfolio, and the capital allocation line.
Using the latest version of yfinance, get closing prices for WMT since January 2020. Compute returns as percent changes.
Get monthly Fama-French factors from French’s data library using pandas datareader. Divide the Fama-French data by 100 to put it in decimal format. Reconcile the date formats of the Fama-French and Yahoo data and merge them.
Compute excess returns for WMT using RF from the Fama-French data. Regress the excess returns on Mkt-RF. Create a scatter plot with regression line.
Workers from over 90% of the companies we surveyed reported regular use of personal AI tools for work tasks. In fact, almost every single person used an LLM in some form for their work.
We use tools to teach concepts
But we can’t just teach tools: tomorrow’s AI tools will be different from today’s.
In March 2025, Y Combinator CEO Garry Tan and managing partner Jared Friedman stated that
for roughly a quarter of the startups in their Winter 2025 cohort, 95% of the codebase was written by AI.
Browser-based apps can be deployed:
Create a retirement planning app that allows a user to enter (i) current account balance, (ii) annual savings (constant), (iii) years to retirement, (iv) mean and standard deviation of annual returns.
Simulate normally distributed annual returns for 1,000 lifetimes and present an analysis, including a density plot of final balances and percentiles. Highlight the mean and median on the density plot.
Create a streamlit app in which the user inputs a ticker, and the app calculates a beta for the stock and produces the scatterplot as done for WMT before and also implements the following:
My ngrok access key is stored as a secret key. Use it to deploy the app.
Add the following functionality to the streamlit app:
Use pandas datareader to get the most recent three-month T-bill rate from FRED. Assume a 6% market risk premium. Use the risk-free rate and the estimated beta to compute the cost of equity capital.
Create a Word doc containing the scatterplot and a table displaying the cost of capital calculation. Provide a download link. Deploy on ngrok.
Add the following functionality to the Streamlit app:
Use my OpenAI API key stored as a secret key. Send the ticker and all elements of the cost of capital calculation to GPT 4.1 and ask for a discussion of the analysis and its meaning.
Include the analysis from GPT 4.1 in the Word doc with the scatterplot and cost-of-capital table. Deploy on ngrok.
A system prompt is text that is sent to the LLM along with each user prompt. It contains information and instructions for the LLM.
Create a custom chatbot as a streamlit app. The app should accept a user prompt and display a response.
Use my OpenAI API key that is stored as a secret key. Route prompts to GPT 4.1.
For the system prompt, use “Reply in pig Latin.” Deploy on ngrok.
From Anthropic:
Through data providers, Claude has real-time access to comprehensive financial information:
Fortune, 9-14-2025:
PromptQL, an enterprise AI platform created by San Francisco-based developer tooling company Hasura, is doling out $900-per-hour wages to its engineers tasked with building and deploying AI agents to analyze internal company data using large language models (LLMs).
Tanmai Gopal, PromptQL’s cofounder and CEO, said “MBA types … are very strategic thinkers, and they’re smart people, but they don’t have an intuition for what AI can do.”
May require chatting at this stage to clarify user’s request
Response could be error message
If so, agent should send to LLM for new SQL code
Create a LaTeX doc in the article style with Hello World as the title. Compile to pdf.
QUESTIONS?