Build a Traditional Medicine Knowledge Graph Agent with Claude Agents SDK and Together AI
At the Global Technical Meeting on AI in Traditional Medicine organized in Delhi, experts emphasized standardizing Ayush knowledge representations. In this challenge, you will construct a specialized Knowledge Graph and reasoning agent using the Claude Agents SDK and Together AI high-throughput inference runtime. The agent will ingest traditional Ayurvedic and Herbal formulations, build a structured RDF/Property Graph, and evaluate herb-drug interactions when traditional treatments are combined with standard modern pharmacotherapy.
What you are building
The core problem, expected build, and operating context for this challenge.
Orchestrate traditional medicine knowledge graph inference agents using Claude Agents SDK powered by Together AI runtime.
How work is evaluated
Evaluates accuracy of herb-drug interaction flags against standard clinical reference databases.
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.
interaction_warning_test
Verifies that CNS additive effect between Ashwagandha and Zolpidem is flagged.
This dimension contributes its full weight only when the submission satisfies the requirement. Partial credit is not awarded.
precision
Precision of herb-drug contraindication flags. • target: 0.95 • 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 extended thinking and tool selection loops with Claude Agents SDK.
Utilize Together AI for high-speed open-weights LLM inference.
Construct property graphs mapping herbs, active phytochemicals, and target metabolic pathways.
Detect potential contraindications and severe herb-drug interaction warnings.
Reference links and supporting material
Curated database of traditional Ayurvedic herb active compounds cross-referenced with PubChem and DrugBank IDs.
Challenge setup
Evaluates accuracy of herb-drug interaction flags against standard clinical reference databases.
Challenge input
JSON with traditional_herbs, modern_drugs, patient_condition.
Claude Agents SDK
Required SDK for complex reasoning and tool usage.
Together AI
Required runtime for fast open-weights LLM execution.
Evaluated output
JSON with interaction_detected, severity_level, clinical_note.
- Verifies that CNS additive effect between Ashwagandha and Zolpidem is flagged.
- Precision of herb-drug contraindication flags. • target: 0.95 • range: 0-1
- Benchmark: Biomedical RAG Interaction Benchmark
- Precision target: 0.95
- 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
- RAI · Required
- Zed · Optional
- Together AI · Optional
- Together AI · Optional
Evaluation contract
- interaction_warning_test · Weight 1
- precision · Weight 1
Metrics
- precision · Target 0.95
Benchmarks
- Biomedical RAG Interaction Benchmark
- biotech
Reference cases
- 1 public reference case
Workflow tasks
- Task 1
Input: JSON with traditional_herbs, modern_drugs, patient_condition.
Output: JSON with interaction_detected, severity_level, clinical_note.
Execution harness: Python
Evaluation tools
- Claude Agents SDK · Agent Framework
- Together AI · Inference Provider
Recipe freeze marker
No recipe freeze time is recorded.