Repository navigation
Solution
Nguyen Le edited this page Nov 16, 2026
·
6 revisions
PortfolioGuard AI is a pause button before Buy or Sell: five agents check the move, explain the likely cause and tell you what it means for your profile, in seconds.
flowchart LR
A["Connect Claude<br/>persona: low / mid / high"] --> B["Choose account<br/>TFSA · RRSP · Non-registered"]
B --> C["Refresh market<br/>second snapshot loads"]
C --> D["Tap Buy or Sell"]
D --> E["Pop-up: short verdict<br/>and why"]
E --> F["Chat: full evidence<br/>and execution trace"]
| Principle | How it is built |
|---|---|
| Numbers never come from a model | Lambda calculates every figure and verdict; models route, read headlines and write explanations that are validated (no digits, only cited evidence) |
| Personal, not generic | The same sale of NVDA is panic for a high-risk long-horizon investor and a reasoned decision for a low-risk, short-horizon one |
| Position-aware | Selling a holding that is a large share of your portfolio (for example NVDA at about 42% of Amira's) is held to a higher evidence bar, and the warning names the concentration risk |
| Data-aware | A move is judged against that ticker's own 60-day volatility and against the market-wide move, not a flat percentage |
| Inspectable | Each trace step shows the AWS AgentCore runtime that ran it |
| Fast | Market and Classifier preload on refresh, so a click finds answers waiting |
| Bring your own AI | Your everyday assistant connects through MCP; nothing to copy and paste |
| Runs offline too |
scripts/demo_offline.sh runs the same calculations and agents locally on the simulated data, with no AWS calls |
Simulation only: no brokerage, no real trades, no predictions.