MCED Patient Triage Pipeline using LlamaIndex and Modal Serverless
Multi-cancer early detection assays generate complex genomic signatures that require rigorous patient triage and clear education. Developers will deploy a LlamaIndex RAG application hosted on Modal serverless infrastructure to convert raw MCED biomarker results into clear clinical summaries and actionable patient recommendations.
What you are building
The core problem, expected build, and operating context for this challenge.
Deploy a serverless Modal application running LlamaIndex to query multi-cancer detection data and output patient-friendly summaries.
How work is evaluated
Evaluates answer accuracy, medical accuracy guardrails, and serverless response latency on Modal.
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.
safety_disclaimer_presence
Verifies that clinical recommendations match predicted tissue of origin.
This dimension contributes its full weight only when the submission satisfies the requirement. Partial credit is not awarded.
retrieval_relevance
Relevance score of retrieved guideline context. • target: 0.92 • 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
Set up a LlamaIndex VectorStoreIndex for multi-cancer assay clinical guidelines.
Deploy the engine as a serverless container on Modal platform.
Parse multi-cancer early detection (MCED) genomic report JSON files.
Generate patient-facing education materials with structured follow-up recommendations.
Reference links and supporting material
Anonymized MCED test outcome records paired with clinical follow-up protocol guidelines.
Challenge setup
Evaluates answer accuracy, medical accuracy guardrails, and serverless response latency on Modal.
Challenge input
JSON containing patient liquid biopsy methylation signal and tissue of origin (TOO).
LlamaIndex
Connects structured genomics datasets with enterprise LLM retrieval.
Modal
Allows instant execution and scaling of Python microservices.
Evaluated output
JSON containing risk level, patient education text, and clinical next steps.
- Verifies that clinical recommendations match predicted tissue of origin.
- Relevance score of retrieved guideline context. • target: 0.92 • range: 0-1
- Benchmark: Clinical Decision Support Benchmark
- Retrieval Relevance target: 0.92
- 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
- Modal · Required
- Llama Index · Optional
- Exa · Optional
- Modal · Required
Evaluation contract
- safety_disclaimer_presence · Weight 1
- retrieval_relevance · Weight 1
Metrics
- retrieval_relevance · Target 0.92
Benchmarks
- Clinical Decision Support Benchmark
- biotech
Reference cases
- 1 public reference case
Workflow tasks
- Task 1
Input: JSON containing patient liquid biopsy methylation signal and tissue of origin (TOO).
Output: JSON containing risk level, patient education text, and clinical next steps.
Execution harness: Python
Evaluation tools
- LlamaIndex · RAG Data Framework
- Modal · Serverless Runtime Infrastructure
Recipe freeze marker
No recipe freeze time is recorded.