PRI Voice Agent

Modular voice first agent framework with extensible skill routing

PRI Voice Agent — modular voice assistant with intent-based skill routing; flagship portfolio analytics (Monte Carlo GBM + sentiment + Entry/Hold/Exit) plus extensible modes for interactive audio demos.

Portfolio mode

• Run Monte Carlo / GBM simulations for stock price paths
• Optionally summarize news sentiment (via NewsAPI)
• Provide a simple Entry / Hold / Exit suggestion
• Generate a simulation video (.webm) if ffmpeg is available

Scope note: This repository is for the PRI Voice Agent (portfolio assistant).
It is not associated with PRI OS, the separate self-optimizing operating system project.

Disclaimer: This project is for educational and demo purposes only and is not financial advice.

---

Contents

• Demo / Presentation
• Features
• How it Works
• Project Structure
• Setup
  • 1) Create Environment
  • 2) Install Dependencies
  • 3) Configure Environment Variables
• Run
  • Voice Mode
  • CLI Mode
• Configuration
• Outputs
• Troubleshooting
• Roadmap
• License

---

Demo / Presentation

• Project deck: docs/Portfolio-PRI.pdf

---

Features

✅ Voice-first interaction

• Wake word: “hey”
• Records your question for a configurable time window
• Text-to-speech via a practical robot voice (macOS say by default)

✅ Portfolio simulation (GBM / Monte Carlo)

• Uses recent historical price data to estimate drift + volatility
• Simulates multiple price paths over a horizon (e.g., 30–60 days)
• Outputs:
  • Expected return estimate
  • Range of possible prices (min/max from simulation endpoints)

✅ Entry / Hold / Exit signal

Simple decision logic combining:

• Expected return (from simulations)
• News sentiment (optional)

✅ Video simulation output (optional)

• Produces frame-by-frame plots
• Converts them into a video using ffmpeg
• Saves video to data/<TICKER>_simulation.webm

---

How it Works (High-level)

1. Wake word detection (voice mode)
2. Speech recording for a fixed duration
3. Text intent parsing
   • portfolio mode triggers: simulate/stock/portfolio
   • fuzzy matching helps map misheard names (e.g. “matter” → “meta”)
4. Market data fetch
   • primary: yfinance
   • fallback logic can use cached data (if available)
5. Simulation
   • GBM-based Monte Carlo runs trials × days
6. Sentiment (optional)
   • fetch headlines via NewsAPI
   • classify sentiment
7. Decision signal
   • Entry / Hold / Exit
8. Outputs
   • summary text
   • plot + optional video

---

Project Structure

PRI/
├─ src/
│  ├─ app.py              # Core logic: simulation, sentiment, CLI commands
│  └─ continuous.py       # Voice mode: wake word, recording, routing to app.py
├─ docs/
│  └─ Portfolio-PRI.pdf   # Presentation deck
├─ .env.example           # Environment variable template (no secrets)
├─ .gitignore
└─ README.md