Positioning¶

Use this page when you need canonical product language for README copy, docs updates, slides, demos, or release notes.

Canonical Product Identity¶

Preferred short label:

• EHR AI platform

Preferred one-sentence description:

OneEHR is a unified Python platform for longitudinal EHR experiments across ML, DL, and LLM agents.

Preferred expanded description:

OneEHR is a unified Python platform for longitudinal EHR experiments. It provides shared infrastructure for preprocessing, modeling, analysis, and reproducible evaluation across AI agents, LLM systems, and conventional ML/DL models on one shared run contract — the first toolkit bridging classical machine learning, deep learning, and agentic AI for clinical prediction.

Core Claims¶

Use these ideas repeatedly and consistently:

• Unified ML/DL/LLM comparison — the first platform that evaluates classical models, deep learning, and LLM agents on the same contract.
• 25 model architectures — tabular ML, recurrent, transformer, Mamba, EHR-specialised, and survival models.
• Built-in dataset support — converters for MIMIC-III, MIMIC-IV, and eICU with standard clinical tasks.
• Medical code ontologies — ICD-9/10 mapping, CCS grouping, ATC drug classification for dimensionality reduction.
• Survival analysis — Cox PH and discrete-time models with concordance index and Kaplan-Meier visualization.
• Statistical rigor — bootstrap confidence intervals, DeLong test, McNemar test, BH FDR correction as defaults, not afterthoughts.
• Fairness-aware — demographic parity, equalized odds, predictive parity, SMD integrated into the standard workflow.
• Interpretability — SHAP, LIME, integrated gradients, permutation importance, attention visualization.
• Publication-ready outputs — ROC, PR, calibration, DCA, forest plots, KM curves with Nature/Lancet style presets.
• Reproducibility by design — TOML config = complete experiment specification, Parquet + JSON artifacts.

Preferred Terms¶

• platform
• AI agents
• LLM systems
• conventional ML/DL models
• shared run contract
• cross-system evaluation
• dataset converters
• medical code ontologies
• survival analysis
• standardized EHR tables
• reproducible artifacts

Terms To Avoid¶

Do not use these as top-level positioning language:

• toolkit (use platform)
• library
• all-in-one
• task-first
• artifact-first
• single-LLM
• multi-agent medical framework
• infra platform
• ML/DL baselines

Competitive Positioning¶

Reusable Copy Blocks¶

Homepage or hero eyebrow:

• EHR AI platform

Short intro:

• OneEHR is a unified Python platform for longitudinal EHR experiments across ML, DL, and LLM agents.

Medium intro:

• OneEHR is a unified Python platform for longitudinal EHR experiments. It provides 25 model architectures, built-in dataset converters for MIMIC and eICU, medical code ontologies, survival analysis, and publication-quality visualization — all on one shared run contract.

Long intro:

• OneEHR is a unified Python platform for longitudinal EHR experiments across ML, DL, and LLM agents. It provides shared infrastructure for preprocessing, modeling, analysis, and reproducible evaluation on one shared run contract — the first toolkit bridging classical machine learning, deep learning, and agentic AI for clinical prediction. With 25 model architectures, built-in converters for MIMIC-III/IV and eICU, ICD/CCS/ATC ontologies, survival analysis, fairness analysis, and publication-quality visualization with Nature/Lancet style presets.

Scope Boundaries¶

Prefer:

• AI agents, LLM systems, and conventional ML/DL models

The canonical abstraction levels for OneEHR:

• product identity: platform
• system scope: AI agents, LLM systems, conventional ML/DL models
• architecture: shared run contract
• data: MIMIC-III/IV, eICU, ICD-9/10, CCS, ATC