← Back to APIs

LLM

Safe Medical Language Model

OpenAI Health Bench 65 score. Citation-backed responses, GDPR-ready deployment, <2% hallucination rate.

Why Safety Matters in Medical AI

Generic LLMs Are Dangerous for Healthcare

  • Hallucinate drug names and dosages
  • Provide medical advice without citations
  • Can't say "I don't know" on uncertain topics
  • Regulatory risk (GDPR, HIPAA, FDA)

Persly LLM Safety Features

  • 100% citation coverage (optional mode)
  • Says "I don't know" when uncertain
  • GDPR-ready, SOC 2 Type II compliant
  • Sentence-level source attribution

Citation-Backed Response Example

Query: "What are the contraindications for metformin?"
Persly LLM Response:

Metformin is contraindicated in patients with severe renal impairment (eGFR <30 mL/min/1.73m²)[1], acute or chronic metabolic acidosis[2], and in patients undergoing radiologic studies involving iodinated contrast materials[3].

[1] FDA - Metformin Prescribing Information (https://www.fda.gov/...)

[2] American Diabetes Association Guidelines 2025 (https://diabetesjournals.org/...)

[3] FDA - Metformin Safety Information (https://www.accessdata.fda.gov/drugsatfda_docs/label/...)

Enable citation mode (100% coverage)

When disabled, responses are faster but may lack sources. Enable for regulatory compliance.

Model Specifications

ModelPersly V1
OpenAI Health Bench65/100
Context Window128K tokens
Citation Coverage100% (optional mode)
Hallucination Rate<2% (with citation mode)
ComplianceGDPR-ready, SOC 2 Type II
LatencyCitation mode: ~15s / Standard mode: ~5s

Code Example

curl https://api.persly.ai/v1/llm/generate \
  -H "Authorization: Bearer YOUR_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "prompt": "What are the contraindications for metformin?",
    "citation_mode": true,
    "max_tokens": 500,
    "temperature": 0.3
  }'

Safety & Accuracy Benchmarks

65/100
OpenAI Health Bench

Medical knowledge accuracy across 100 questions

100%
Citation Coverage

Every sentence has a verifiable source (citation mode)

<2%
Hallucination Rate

False medical facts (citation mode enabled)

98%
"I Don't Know" Accuracy

Correctly identifies uncertain/out-of-scope questions

OpenAI Health Bench Score Comparison

ModelScore
Persly V165
GPT-550
Gemini 2.538

* OpenAI Health Bench: https://github.com/openai/simple-evals/tree/main/health_bench * Hallucination rate measured on 500 medical Q&A pairs

Use Cases

Medical Chatbots

Power patient-facing chatbots with safe, citation-backed answers

Clinical Summarization

Summarize patient records, research papers, clinical notes

Medical Q&A

Answer medical questions with verifiable sources

Drug Information

Provide drug interactions, contraindications, dosing info

FAQ

Is Persly LLM HIPAA compliant?

We're GDPR-ready and SOC 2 Type II compliant. For HIPAA, contact us for BAA (Business Associate Agreement) options.

What's the difference between citation mode on vs off?

Citation mode: 100% coverage, <2% hallucination, slower (~15s). Standard mode: Faster (~5s), may lack sources, ~5% hallucination.

Can I use Persly LLM for non-medical use cases?

Yes, but it's optimized for healthcare. For general use, consider generic LLMs (OpenAI, Anthropic).

What languages are supported?

100+ languages. Citations include both local and international sources.

Ready to Build with Persly?

Let's discuss how our APIs can power your healthcare product