CareLoop

Turns a clinician's spoken visit into a grounded care plan that families can ask Alexa+ about, with no invented advice and no spoken doses.

Entry for the Amazon Developer Hackathon (Alexa+ track + Open Source mini challenge). Adult care plans only; all data is synthetic.

Core Rules (enforced in code + CI)

• Extractive by construction: the model returns only span offsets; every string is a verbatim clinician quote or a fixed constant, and non-verbatim quotes are rejected.

• Dose-free voice: every quantity is masked to "your prescribed amount" before speech; full doses appear only on the authenticated companion app. Any leak is a kill criterion, re-checked independently by preflight/.

Related MCP server: Hearth v2

Key Features

• MCP server: 12 tools over Streamable HTTP, with schemas, plain-language errors and generated contracts.

• Grounded capture: offline rules engine always runs; optional Bedrock two-pass (extract, then verify).

• Voice safety: dose masking, crisis-to-helpline redirect, adult-only gate, prompt-injection refusal.

• Safe dialogs: destructive changes are read back and need an explicit yes; ambiguity triggers a question, never a guess.

• Adaptive delivery: "slower / louder / softer / shorter" re-speaks the answer with no tool call or write.

• Speaker scope: the service decides access from the household boundary (simulated profile, not biometrics).

• Two devices, one protocol: browser Device Lab and headless driver, diffed turn by turn.

• Privacy: encrypted SQLite per household, hashed transcripts, tamper-evident ledger, crypto-shred receipt.

• Auth: OAuth 2.1 with PKCE (S256), RFC 9728 metadata, rotating refresh tokens.

• Accessibility: 48 px targets, WCAG-AA contrast, reduced motion, English/Hindi/Punjabi.

Four Surfaces, One Origin

• /simulator: Alexa+ Device Lab.

• /card: single self-contained 72.7 kB HTML file.

• /companion: Angular app to paste a visit, view full doses, confirm, mark done, export and erase.

• /judge and /: policy self-test, ledger verify, latency, and docs site.

• All served by one stdlib-only Python process, offline.

Quickstart

Docker

docker build -t careloop .
docker run --rm -p 127.0.0.1:8080:8080 -v careloop-data:/data careloop
# → http://127.0.0.1:8080/simulator

Local (Python 3.10+, Node 20+)

python -m pip install -r requirements.txt && npm install
npm run build && npm run sync
python scripts/serve.py   # http://localhost:8080/simulator

• Demo household: CASA01, PIN 481542.

• Secrets are env-only (CARELOOP_AUTH, CARELOOP_DB_ROOT, CARELOOP_PUBLIC_URL, etc.); .env is gitignored.

Verified Results (measured 2026-10-23)

• Python tests: 681 passed

• Frontend tests: 231 passed across 22 files

• Eval: 0.0% hallucination, 0.0% dose leak, 3/3 refusals, F1 0.965

• Voice eval: 100% on model's own samples, 0/12 out-of-domain answered (deliberate)

• Device scenarios: 10 scenarios, 221 checks, 0 dose leaks

• Device comparison: both devices agree, 0 divergences

• Preflight: 39 checks, 35 pass, 0 fail, 4 skip

• Smoke test: PASS

• Real published prose: 142 docs, 2,291 sentences, 0 hallucinations, 0 leaks

Repo & Licence

• careloop/ server, frontend/ (5 workspaces), alexa/ skill package, artifacts/, tests/, preflight/, scripts/, evidence/, Dockerfile.

• MIT licence; stdlib-only runtime with two optional packages (cryptography, boto3).

• No web fonts, icon libraries, audio files or third-party brand assets.

Screenshots