Tether
Tether turns a missed follow-up into a phone call instead of a hospital bed.
Core trick
Rule engine over home readings and labs in SQL flags missed follow-ups, then Cortex Claude drafts the next action for clinician approval.
Tags
Description
Tether turns a missed follow-up into a phone call instead of a hospital bed.
Chronic disease is managed at home, and almost no one is checking in. A primary care doctor would need 26.7 hours a day to deliver all recommended care, so follow-up calls are the first thing to go, and 3.5 million adult hospital stays a year are potentially preventable.
Tether keeps patients with hypertension, COPD and type 2 diabetes tethered to their care plan. It does the follow-up a practice cannot staff, and hands the clinician only the decisions. It watches home readings, labs and refills in Snowflake, applies guideline rules (2025 AHA/ACC, GOLD 2026, ADA 2026) plus the doctor's signed AI Plan from the visit note, and uses Claude through Cortex to draft the next step: a patient check-in, an AI follow-up call, or a request to the prescriber or pharmacy. Missed check-ins climb a follow-up ladder, and red flags go straight to the top of the queue. A clinician approves every outbound action in one click. Tether does not diagnose and does not choose drugs or doses.
Built on Snowflake (a SQL rules engine, role-separated approval, Cortex AI_COMPLETE with Claude, and a Cortex Agent over a semantic view for plain-language questions). All data is synthetic.
README excerpt
# Tether Tether keeps patients with hypertension, COPD and type 2 diabetes tethered to their care plan. It watches home readings and labs in Snowflake, applies guideline rules (2025 AHA/ACC, GOLD 2026, ADA 2026), and uses Claude to draft the next step: a patient check-in, an AI follow-up call, or a request to the prescriber or pharmacy. **A clinician approves every outbound action.** Tether does not diagnose and does not choose drugs or doses. Built at the Intelligence at Work Hackathon (Snowflake x Anthropic), Oct 23, 2026. All code and data in this repo were created during the event. Data is synthetic; there is no real patient information. ## What's here | Path | What it is | | --- | --- | | `docs/RUNBOOK.md` | The runbook: pathways, thresholds and sources, safety checks, follow-up ladder, AI check-in calls, approval gate, test scenarios, hackathon checklist | | `docs/DEMO-VIDEO-SCRIPT.md` | Shot-by-shot script and SQL for the 1-minute demo video | | `docs/PITCH.md` | Two slides of pitch wording: clinician time, hospital admissions, life expectancy, lost workdays, families, with sources | | `data/soap_notes.csv` | 45 synthetic SOAP notes; 14 end with a clinician-signed AI Plan | | `data/gen_notes.py` | Generator for the notes | | `sql/care_rules_snowflake.sql` | Snowflake schema, editable rules table, 15 synthetic patients and their notes, AI Plan parsing with safety checks, rule evaluation view, Claude drafting through Cortex AI_COMPLETE, clinician approval procedure