Build a LlamaIndex AYUSH DPI Traditional Medicine RAG Knowledge Engine
Integrating traditional medicine into digital public health stacks requires verifiable retrieval over classical Ayurvedic texts and modern clinical trials. Build a LlamaIndex hybrid RAG pipeline with dynamic metadata filtering to answer clinical queries with 95% source attribution accuracy.
What you are building
Develop a LlamaIndex RAG system over classical AYUSH medical texts featuring metadata filters and chunk-level citation tracking.
How work is evaluated
Evaluates query retrieval recall and exact citation match precision.
Datasets
Public datasets, plus the private datasets you have access to.
How submissions are scored
attribution_check
Verifies source attribution score meets threshold
This dimension contributes its full weight only when the submission satisfies the requirement. Partial credit is not awarded.
attribution_score
Accuracy ratio of cited source documents to generated facts • target: 0.96 • range: 0-1
This dimension contributes its full weight only when the submission satisfies the requirement. Partial credit is not awarded.
Learning goals
Construct LlamaIndex vector and keyword hybrid indices over classical medical texts
Implement dynamic metadata filtering by herb taxonomy and formulation type
Enforce precise source attribution and verbatim chunk reference extraction
Deploy sub-node query engine pipelines for clinical decision support
Resources
Corpus of digitized WHO and AYUSH traditional medicine research reports and clinical trial metadata.
Challenge setup
Evaluates query retrieval recall and exact citation match precision.
Challenge input
JSON query string regarding botanical formulation
LlamaIndex
RAG index creation and retrieval orchestration
Evaluated output
JSON answer string with citation_ids array
- Verifies source attribution score meets threshold
- Accuracy ratio of cited source documents to generated facts • target: 0.96 • range: 0-1
- Benchmark: Medical RAG Benchmark
- Attribution Score target: 0.96
- 1 public reference case
- Python execution harness
- Python sandbox (unavailable on Versalist)
Shows the first task, up to three tools, and up to three evidence items.
View technical recipe
Configured tools
- Llama Index · Required
- Exa · Optional
- Langfuse · Optional
- Langfuse · Optional
Evaluation contract
- attribution_check · Weight 1
- attribution_score · Weight 1
Metrics
- attribution_score · Target 0.96
Benchmarks
- Medical RAG Benchmark
- biotech
Reference cases
- 1 public reference case
Workflow tasks
- Task 1
Input: JSON query string regarding botanical formulation
Output: JSON answer string with citation_ids array
Execution harness: Python
Evaluation tools
- LlamaIndex · Agent Framework
Recipe freeze marker
No recipe freeze time is recorded.