Inspiration
Self-directed investors often don't know what they really own once ETFs overlap, or how much of their risk sits in one position. Most AI investing demos recommend a trade; an order-execution-only broker may not. We wanted a tool that shows what you own and what you risk, before the order, without ever giving advice.
What it does
ÉPICENTRE: Know what you own. Know what you risk. A portfolio and order preview for self-directed investors.
- See through your ETFs: direct and fund holdings combined, using dated issuer disclosures with coverage shown.
- Money vs. Risk: for Robert, XIU holds 45 % of his money but about 71 % of his modeled portfolio risk.
- Preview an order: replay the same market history with and without the trade (daily adjusted closes in CAD through 12 Nov 2026) against the investor's own limits, then open the evidence behind the lowest balance.
- Four cases: Robert (capital preservation), Mateo (values screen and disclosure gaps), Amira (15 % crypto vs. a 10 % cap, NVIDIA exposure across direct and ETF positions) and "Too Canadian?" (country exposure and drawdown).
- Use my portfolio: in the AWS workspace, enter your own positions and set optional limits in plain language.
Everything is simulated: no transactions and no buy, sell or hold.
How we built it
- The application computes allocation, risk and historical losses: code does every number.
- Amazon Bedrock Guardrails screen the investor's question.
- Claude on Amazon Bedrock uses an
open_evidencetool to pick up to three existing evidence cards; the app validates the selected IDs and shows the calculated result with its sources. - Python and FastAPI, deployed with Docker on AWS behind Amazon CloudFront; the public demo runs on GitHub Pages with saved cases.
- Data: the organizers' starter pack (daily prices, FX and rates) and dated ETF issuer disclosures.
Challenges we ran into
Keeping the AI explaining without ever producing a number, making ETF look-through honest when disclosures are incomplete, and staying clearly on the "no advice" side.
Accomplishments that we're proud of
Every result traces back to its data and calculation, the risk view surfaces concentration that the money view hides, and the whole thing works for a real investor's own portfolio.
What we learned
Investors trust results they can trace. Let the model choose and explain, and let code calculate.
What's next for ÉPICENTRE
An order preview beside the order ticket of a self-directed brokerage, at the client's click and subject to compliance review.
Educational simulation. Not financial advice.
Built With
- amazon-bedrock
- amazon-cloudfront
- amazon-cloudwatch
- amazon-ec2
- amazon-web-services
- aws-iam
- bedrock-guardrails
- claude
- docker
- fastapi
- github-actions
- javascript
- pytest
- python