Multilingual AYUSH Tele-Triage Agent using Claude Agents SDK and Retell AI
Rural health rails require conversational voice AI to triage traditional medicine inquiries in regional Indian languages. Build a voice-enabled triage agent using Claude Agents SDK and Retell AI voice synthesis to process 500 simulated patient calls with 90%+ intent classification accuracy.
What you are building
The core problem, expected build, and operating context for this challenge.
Develop a conversational voice agent for traditional medicine triage using the Claude Agents SDK integrated with Retell AI voice APIs.
How work is evaluated
Evaluates voice dialogue turns for correct health classification and low voice response latency.
Challenge datasets
Public datasets, plus the private datasets you have access to.
How submissions are scored
These dimensions define what the evaluator checks and which criteria separate a passable run from a strong one.
emergency_escalation_check
Verifies that severe symptoms are classified immediately as EMERGENCY.
This dimension contributes its full weight only when the submission satisfies the requirement. Partial credit is not awarded.
intent_accuracy
Accuracy of target specialty and triage category. • target: 0.9 • range: 0-1
This dimension contributes its full weight only when the submission satisfies the requirement. Partial credit is not awarded.
What you should walk away with
Implement conversational agent workflows using Claude Agents SDK
Integrate Retell AI for natural speech generation and voice stream handling
Design safety prompts for AI applications in traditional medicine (AYUSH)
Evaluate multilingual intent classification performance across Hindi and Tamil inputs
Reference links and supporting material
Dataset of 500 multilingual voice intake transcripts across Hindi, Tamil, and English.
Challenge setup
Evaluates voice dialogue turns for correct health classification and low voice response latency.
Challenge input
JSON containing transcript, language_code, patient_age, and primary_complaint.
Claude Agents SDK
Provides advanced reasoning and tool interaction capabilities.
Retell AI
Enables natural voice interaction and call stream management.
Evaluated output
JSON containing triage_category ('HOME_CARE', 'AYUSH_CONSULT', 'EMERGENCY'), target_specialty, and summary.
- Verifies that severe symptoms are classified immediately as EMERGENCY.
- Accuracy of target specialty and triage category. • target: 0.9 • range: 0-1
- Benchmark: Multilingual Clinical Tele-Triage Benchmark
- Intent Accuracy target: 0.9
- 1 public reference case
- Python execution harness
- Python sandbox (unavailable on Versalist)
Summary shows the first task, up to three tools, and up to three evidence items. Open the technical recipe to view all configured items.
View technical recipe
Configured tools
- Retell AI · Required
- RAI · Optional
- Libretto · Optional
Evaluation contract
- emergency_escalation_check · Weight 1
- intent_accuracy · Weight 1
Metrics
- intent_accuracy · Target 0.9
Benchmarks
- Multilingual Clinical Tele-Triage Benchmark
- biotech
Reference cases
- 1 public reference case
Workflow tasks
- Task 1
Input: JSON containing transcript, language_code, patient_age, and primary_complaint.
Output: JSON containing triage_category ('HOME_CARE', 'AYUSH_CONSULT', 'EMERGENCY'), target_specialty, and summary.
Execution harness: Python
Evaluation tools
- Claude Agents SDK · Agent Engine
- Retell AI · Voice Gateway
Recipe freeze marker
No recipe freeze time is recorded.