How apprajauniq/lepto-multiagent-mortality works
This repository implements a seven-stage LLM workflow for ICU mortality prediction, then analyzes and visualizes its results. The README documents the architecture and module responsibilities, but source excerpts were not provided; links below reflect documented workflow and CLI behavior rather than inspected call sites.
Components 24
Inference pipeline
- Pipeline CLI
- Batch orchestration
- Patient formatting
- Agent prompts
- Five specialists
- Intensivist
- Judge
- LLM client
- JSON parsing
Statistical analysis
- Statistics CLI
- Results loading
- Classification metrics
- DeLong tests
- Failure analysis
Results presentation
- Figures CLI
- Figure generation
Configuration
- Configuration
- YAML settings
Other components
- Patient CSV
- Groq API
- Results CSV
- Statistics JSON
- Figures 1–5
- Environment variables
Connections 31
- Patient CSV → Pipeline CLI: provides input
- Pipeline CLI → Batch orchestration: runs
- Pipeline CLI → Configuration: loads settings
- Batch orchestration → Patient formatting: formats records
- Batch orchestration → Five specialists: runs on records
- Batch orchestration → Intensivist: runs after specialists
- Batch orchestration → Judge: runs after integration
- Batch orchestration → Results CSV: writes results
- Patient formatting → Five specialists: supplies patient text
- Five specialists → Intensivist: supplies predictions
- Five specialists → Judge: supplies predictions
- Intensivist → Judge: supplies prediction
- Five specialists → Agent prompts: uses prompts
- Intensivist → Agent prompts: uses prompts
- Judge → Agent prompts: uses prompts
- Five specialists → LLM client: requests inference
- Intensivist → LLM client: requests inference
- Judge → LLM client: requests inference
- LLM client → Groq API: calls API
- LLM client → JSON parsing: parses response
- Statistics CLI → Results loading: loads results
- Results CSV → Results loading: provides data
- Statistics CLI → Classification metrics: computes metrics
- Statistics CLI → DeLong tests: runs paired tests
- Statistics CLI → Failure analysis: analyzes dissent
- Statistics CLI → Statistics JSON: writes summary
- Figures CLI → Figure generation: generates figures
- Results CSV → Figures CLI: provides data
- Figures CLI → Figures 1–5: writes PNGs
- Configuration → YAML settings: reads settings
- Configuration → Environment variables: reads secrets