pocketgull-steeep-quality

PocketGull STEEEP Quality (Healthcare Quality Sentinel)

pocketgull-steeep-quality is a specialized clinical alignment adapter developed by Phil Gear as part of the PocketGull Clinical Intelligence Fleet. Fine-tuned on google/gemma-2-2b-it, it addresses critical clinical challenges in Institute of Medicine (IOM) Quality Auditing.

---

🏥 Key Clinical Capabilities

1. Clinical Safety & Harm Prevention: Flags potential diagnostic and therapeutic errors before care delivery.
2. Health Equity Auditing: Evaluates clinical algorithms for disparate impact and demographic bias.
3. Efficiency & Waste Reduction: Identifies redundant laboratory orders and non-indicated diagnostic scans.
4. Patient-Centered Respect: Audits clinical communications for shared decision-making and health literacy alignment.

---

🔬 Empirical Safety Verification (Inspect AI)

Evaluated under the Inspect AI (UK AI Safety Institute) framework across four clinical safety vectors:

---

🚀 Quickstart & Inference

import torch
from transformers import AutoModelForCausalLM, AutoTokenizer
from peft import PeftModel

base_model_id = "google/gemma-2-2b-it"
adapter_id = "philgear/pocketgull-steeep-quality"

tokenizer = AutoTokenizer.from_pretrained(base_model_id)
base_model = AutoModelForCausalLM.from_pretrained(
    base_model_id,
    torch_dtype=torch.bfloat16,
    device_map="auto"
)
model = PeftModel.from_pretrained(base_model, adapter_id)

prompt = "Clinical query regarding Institute of Medicine (IOM) Quality Auditing."
inputs = tokenizer(prompt, return_tensors="pt").to(model.device)
outputs = model.generate(**inputs, max_new_tokens=256)
print(tokenizer.decode(outputs[0], skip_special_tokens=True))

---

🔍 Independent Audit & Reproducibility

# Independent audit command via Inspect AI
pip install inspect-ai peft torch transformers
inspect eval test_clinical_safety.py --model hf/philgear/pocketgull-steeep-quality

---

⚖️ Governance & Statutory Duty of Care Notice

This model adheres to the Digital Duty of Care & Safety by Design (SbD) framework:

• Statutory Compliance: Pre-flight evaluation against the Australian eSafety Commissioner Safety by Design principles.
• Clinical Harm Modeling: Risk profiled under the STRIDE-H (Human Harms) Threat Matrix.
• Non-Diagnostic Notice: This model is designed for clinical workflow validation, triage support, and safety auditing. It does not replace direct physician clinical judgment.

---

📄 Citation & Attribution

@misc{pocketgull_steeep_quality_2026,
  author = {Phil Gear},
  title = {PocketGull STEEEP Quality (Healthcare Quality Sentinel)},
  year = {2026},
  publisher = {Hugging Face},
  howpublished = {\url{https://huggingface.co/philgear/pocketgull-steeep-quality}}
}