4.5 KiB
hncb-fusion-deid-demo
All-AWS demo of the HNCB use case: a reversible PII de-identification round trip through Axway Amplify AI Gateway (Fusion), with the reasoning done by an Amazon Bedrock AgentCore agent. The advisor types a query containing a real name; the cloud only ever sees a token; the real identity is restored on-prem before the answer is shown.
Status: unrun scaffold. This was authored, not executed. Review everything, pin dependency versions, and expect to adjust AgentCore/Fusion specifics to the current CLI/console. Demo-grade, not production-grade.
What it demonstrates
Raw PII (a Chinese name + a Taiwan ROC ID) is detected on ingress, replaced with a
random, format-safe token, and only the tokenized prompt goes to the cloud
agent. The agent calls a tool back "on-prem" with the token, gets a de-identified
evidence package, and writes talking points. Fusion restores the identity on the
way out. The money shot: show the Bedrock request / AgentCore trace live — the
cloud only ever saw CUST_000123, never 王小明.
Trust zones are logical
Everything is one AWS account. The "on-prem" zone is a set of tag-labelled
resources (Zone = on-prem-VPC-A) standing in for HNCB's branch data centre. This
demo proves behaviour and data-flow, not physical data residency — say so on
camera: "in production this zone is the branch DC; here a VPC stands in for it."
Component → service map (their 8 steps)
| Step | Component | Service in this repo |
|---|---|---|
| 1, 8 | Advisor UI | ui/index.html on S3+CloudFront (or local) |
| 2, 3-route, 8 | Fusion AI Gateway (the product) | ECS Fargate — terraform/ecs.tf, config in fusion/POLICY_SETUP.md |
| 2 | PII detector (typed findings, not redaction) | Presidio on Fargate — presidio/ |
| 2, 4, 8 | Token vault (reversible map) | DynamoDB — terraform/main.tf |
| 3, 7 | Cloud agent + model | AgentCore Runtime + Bedrock — agent/ |
| 4 | Tool bridge (Lambda → MCP tool) | AgentCore Gateway — scripts/agentcore_setup.sh |
| 4-6 | On-prem RAG tool + data | Lambda + DynamoDB — lambda_rag/, seed/ |
| all | Observability | AgentCore Observability + CloudWatch |
Repo layout
terraform/ core infra (DynamoDB, RAG Lambda, IAM) + ECS hosting
lambda_rag/ RAG tool: token resolve -> de-identified evidence package
agent/ Strands agent for AgentCore Runtime + tool schema
presidio/ PII detector service (returns typed findings) + Dockerfile
seed/ fake customer (Wang Xiaoming) + seed script
ui/ advisor UI with restored-vs-tokenized split view
scripts/ deploy.sh, agentcore_setup.sh, teardown.sh
fusion/ POLICY_SETUP.md (the console/flow config — the manual part)
Prerequisites
aws-cli configured (creds + region), terraform >= 1.5, docker, python3,
the AgentCore CLI (npm i -g @aws/agentcore), Bedrock model access enabled for
bedrock_model_id, and an Axway Amplify AI Gateway (Fusion) container image you
supply (fusion_image_uri).
Deploy
bash scripts/deploy.sh # infra -> presidio image -> ECS -> seed -> AgentCore
# then: configure Fusion policy (fusion/POLICY_SETUP.md)
# then: point ui/index.html GATEWAY_URL at the Fusion host and open it
Demo script (maps to the 8 steps)
- Advisor UI: submit "請幫我整理王小明最近三個月的理財往來,並給我下次拜訪話術。"
- Fusion detects
王小明+A123456789, tokenizes, logs tokens only. - Show the Bedrock/AgentCore trace — the prompt the cloud saw contains
CUST_000123. 4-6. Agent tool-calls back on-prem; RAG resolves the token, returns a summary. - Agent writes talking points (no PII).
- Fusion restores
王小明; the UI shows the restored answer beside the tokenized view.
Teardown
bash scripts/teardown.sh # stop paying for Fargate / AgentCore
Honest caveats
- zh-TW detection is demo-narrow. Presidio here is tuned to the scripted entities; a smooth run is not evidence of production zh-TW recall — that remains the real-engagement risk.
- Per-request randomization ("different each time") lives in the Fusion mint
step (
fusion/POLICY_SETUP.md); confirm it satisfies HNCB's requirement. - The agentic token-resolution loop (agent tool call → on-prem RAG resolves the token) is custom orchestration — it is not turnkey on any gateway, which is exactly where Fusion's orchestration depth is the argument.