TL;DR
I run a robo advisory which offers basic portfolio construction/management, financial advice and rebalancing. I’d like to swap its current operation for one in which each client interacts with a personalized AI Financial Advisor to create an individualized portfolio and financial plan. I’ve put together a tentative plan of action on how to accomplish this and I’d love to get some feedback/advice/opinions from the Community. Thank you.
CURRENT OPERATION
Strategic International (SI) is a robo advisory registered with the SEC. It was founded in 2017 and is wholly owned and operated by yours truly.
SI helps clients construct and rebalance efficient investment portfolios as defined by Modern Portfolio Theory; brokerage services, authentication, and custody are handled by a third party (Interactive Brokers).
A client’s Risk Tolerance is quantified using a questionnaire.
A list of exchange symbols is obtained from sources online.
Securities data is gathered. For stocks, these are end-of-day adjusted close prices.
Expected Return and Beta are calculated for each asset in the portfolio, and Expected Return, Return Variance and Standard Deviation are calculated for the portfolio as a whole.
Portfolios are evaluated on their Utility. “Utility" is the difference between the Expected Return adjusted for Risk Tolerance and a “risk-free” rate of return (like a Treasury Bond). A positive number means the portfolio has a risk-adjusted rate of return that exceeds the “risk-free” rate of return.
Automated recommendations are consistent with maximizing the Utility of the portfolio and clients are encouraged to experiment with different securities combinations/quantities to achieve this.
When a client is happy with their portfolio allocations, they can hit “Submit” and SI will do its best to automatically rebalance monthly to maintain allocation percentages.
DESIRED OPERATION
In a nutshell, Strategic International becomes a portal to a Financial Advisor AI agent (Agent) created and updated using Palantir. Clients create accounts and then are asked questions by the Agent about things like Time Horizon, Risk Tolerance, Tax Situation, etc. until it feels it has a good grasp of a client’s financial needs. If a client has an existing portfolio, it offers rebalancing recommendations to efficiently bring it in line with their stated goals. The more a client interacts with the Agent the better it gets to know them, maintaining a memory of all previous interactions.
TENTATIVE PLAN
Create a Data-Connection to existing postgreSQL database.
Create an AI Agent instance using Agent Studio.
Create a Pipeline to give AI Agent access to non-sensitive information on database like stock tickers and historical EOD prices.
Train an AI Agent in finance by feeding it PDF textbooks (assuming Agent is sophisticated enough to “read” and “understand” text with minimal cleaning and investment on my part).
Once the AI Agent demonstrates that it’s able to access prescribed data, ask the right questions and give solid financial advice, the instance will be saved as a template from which to instantiate Personalized Agents for clients. These Personalized Agents have access to a client’s unique financial information like their portfolio composition, income, etc. and may only communicate with that client or myself (to protect personal information). Personalized Agents maintain a memory of past interactions with clients and use this knowledge to better serve them.
Unless a client’s financial situation calls for something different, by default, a Personalized Agent will work with a client to construct an efficient portfolio which is rebalanced monthly to conform with stated allocations.
I’ve copied my post into the Solutions Designer App and had it generate Implementations for each step.
That being said, actually evaluating whether the Plan is an elegant and efficient solution to achieving my goals is not really something I feel I’m qualified to do without first actually building the thing, since the totality of my knowledge of Palantir’s systems is currently limited to about 5 available Tutorials and a handful of Examples.
Unless someone has a different idea, I’m just going to move forward with trying
to implement the generated AIP Architect Walkthrough Guide and see where it leads me.
AIP Architect Walkthrough Overview
Workflow Target: Application
Requirements: Transition Strategic International’s robo advisory to a personalized AI Financial Advisor system using Palantir. The AI will interact with clients to create individualized portfolios and financial plans, maintaining a memory of interactions.
Code Level: Pro-code
Familiar Languages: TypeScript
Data Sources: External system, PostgreSQL database
Special Properties: Time-series data
Data Scale: Less than 1 billion rows
Data Frequency: User-driven
The plan involves creating data connections, AI agents, and APIs to integrate with existing systems and provide personalized financial advice.
Plan your workflow - Use AIP Architect to translate your workflow requirements into a Foundry implementation plan. (AI Financial Advisor System)
Create ‘External System - Strategic International’ - Ingests data from Strategic International’s external systems for financial analysis in Foundry.
Create ‘Magritte Agent - Data Integration’ - Facilitates data transfer from Strategic International’s systems to Foundry using a secure connection. (AIP Agent Playground)
Create ‘Magritte - Data Connector’ - Processes and formats the ingested data for seamless integration into Foundry’s datasets. (Data connection)
Create ‘Dataset - Financial Analysis Data’ - Stores the processed time-series data for use in the AI Financial Advisor system within Foundry. (AIP Agent Playground)
Create ‘Code Repository - PySpark Scripts’ - Hosts PySpark scripts that manage incremental processing and transformation of the raw financial data. (Code repositories)
Create ‘Dataset - Transformed Financial Data’ - Stores the processed and transformed financial data, ready for integration into the AI Financial Advisor system. (AIP Agent Playground)
Create ‘Client’ - Stores information about clients interacting with the AI Financial Advisor. (Ontology Manager)
Create ‘Portfolio’ - Contains details of the client’s investment portfolio. (Ontology Manager)
Create ‘FinancialPlan’ - Details the financial plan created for the client. (Ontology Manager)
Create ‘Interaction’ - Records interactions between the AI and the client. (Ontology Manager)
Create ‘Client-Portfolio Link’ - Link type between Client and Portfolio objects. (Ontology Manager)
Create ‘Client-FinancialPlan Link’ - Link type between Client and FinancialPlan objects. (Ontology Manager)
Create ‘Client-Interaction Link’ - Link type between Client and Interaction objects. (Ontology Manager)
Create ‘Action - Execute Financial Logic’ - Executes specific logic on financial ontology objects, such as updating a client’s portfolio or generating a financial plan. (Ontology Manager)
Create ‘Object - Financial Ontology Objects’ - Represents financial entities like Client, Portfolio, and FinancialPlan, which are manipulated by the Ontology SDK. (Ontology Manager)
Create ‘Ontology SDK - Financial Data Manipulation’ - Facilitates CRUD operations on financial ontology objects like Client, Portfolio, and FinancialPlan, enabling complex logic execution. (Developer Console)
Create ‘Hosted App - AI Financial Advisor’ - Serves as the frontend interface for users to interact with the AI Financial Advisor, enabling them to manage their financial plans and portfolios. (AIP Agent Playground)
Each “Next” in the walkthrough will take you to the next module in the process adding data as you go.
Flow Overview & Files tab help track & display the data you’re pulling in.
Very Organized tool.
