LabLine AI
Inspiration
Critical laboratory results are time-sensitive, but safe communication requires more than simply delivering a value. The recipient must be authorized before disclosure, the approved result must be communicated exactly, the recipient must read it back correctly, and any failure in that closed loop must trigger human escalation.
LabLine AI explores how a safety-bounded CALL-E workflow can make that process explicit, structured and auditable.
What it does
LabLine AI demonstrates a closed-loop workflow for communicating an approved critical laboratory result.
The fixed synthetic demo case is LAB-DEMO-001. The intended recipient is an on-call clinician. Before the result is disclosed, the recipient must provide the exact synthetic authorization code LAB-204.
Only after authorization does LabLine disclose the approved synthetic result:
potassium 6.8 mmol/L
LabLine then requests a read-back.
If the recipient repeats potassium 6.8 mmol/L correctly, the communication closes successfully.
If the recipient instead reads back potassium 6.9 mmol/L, the loop does not close. The workflow fails closed, makes no automatic retry, and requires human escalation.
Safety by design
LabLine is deliberately narrow.
It does not infer or modify the approved laboratory result. It does not disclose the result before authorization. A claimed name or role alone is not sufficient for authorization.
An incorrect read-back does not produce a success state. Instead, the structured result records that the communication remains unresolved and that human escalation is required.
The public structured-output contract is defined in references/result-schema.json.
How it was built
LabLine was implemented as a CALL-E skill with public documentation, examples, safety guidance and a structured result schema.
The contribution was merged into the main CALL-E repository in PR #392.
For judging and reproducibility, the project also includes a deterministic offline simulator. From the repository root, the demo runs with:
cd prototype && python critical_result_simulator.py
The offline demo uses only synthetic data and the Python standard library.
Testing and evidence
The complete offline test suite passes 30/30 tests.
Both structured demo outcomes validate against the public result-schema.json.
The deterministic demo was executed twice and produced byte-for-byte identical output on both runs.
The verified SHA-256 of the demo output is:
7f4fd1cc2c499b8b1c5d11a9dac3082fd9e976ebef21dbb9d32a0da9deabf5ec
The public demo video shows both the successful closed-loop path and the incorrect-read-back fail-closed path.
Demo video
LabLine AI — Closed-Loop Critical Result Demo
Duration: 2:34
Important limitations
The demonstrated workflow is entirely synthetic and offline.
It uses no real patient, clinician, hospital, telephone or laboratory data. It places no telephone call, uses no network connection and requires no CALL-E account or credentials.
The synthetic call_id and run_id values shown in the demo are fixed fixture identifiers, not CALL-E runtime identifiers.
The demo demonstrates deterministic simulated workflow behavior, authorization gating, read-back handling, fail-closed escalation and structured-output compliance.
It does not claim a completed live CALL-E telephone interaction or production clinical deployment.
Log in or sign up for Devpost to join the conversation.