Slides

CoupleVest: Help Couples Agree on Investment

AGI House: AI x Fintech Build Day

Problem:

Virtually ALL couples have to manage joint finances

• Every couple we interviewed have different investment preferences (eg “single stock” vs “broad strategy”)

• CoupleVest helps resolve this by:

  • making ideas concrete: uses LLM for natural language (conviction) -> maths (code) and chart

  • Simulator: uses reasoning LLM to find similar situation in the past and historical data for simulation

Tech

LLM is used to convert natural language ideas into code, eg:

 - “show me Tesla stock price in Euro”
 - “Apple always drops after earnings”

Verifiable AI approach (aka avoid LLM for traiding, only for code<->natural language translation)

• Code is separately explained in natural language using LLM
• Code is then executed to produce charts and tables without any at runtime LLMs

Inspiration: Voyager paper by NVidia et al https://voyager.minedojo.org/

Running

Create .env file with

OPENAI_API_KEY=your_openai_api_key_here
FINNHUB_API_KEY=your_finnhub_api_key_here

OpenAI is used to convert natural language into code.

FinnHub is used to load stocks prices.

Code

Natural language metric creation and execution (using V8 Isolates):

1. vibes-metric-creation.ts

2. execute-metric.ts

Simulator code is here:

simulator.ts

Negotiation Sample

🔊 ✨ Switch ON the AUDIO 🎧 ✅:

output.mp4

Screenshots

Goals

Simulator

Final Agreed Allocation