Publish artifacts for clinical-rag-safety-gateway-20260815

README.md

@@ -0,0 +1,59 @@
+---
+license: cc-by-4.0
+language:
+- en
+pretty_name: Clinical RAG Safety Gateway Synthetic Evaluation Set
+size_categories:
+- n<1K
+task_categories:
+- question-answering
+tags:
+- synthetic
+- healthcare-ai
+- evaluation
+- question-answering
+- sentence-similarity
+- text-classification
+- summarization
+configs:
+- config_name: default
+  data_files:
+  - split: train
+    path: data/train.jsonl
+  - split: test
+    path: data/test.jsonl
+---
+
+# Clinical RAG Safety Gateway Synthetic Dataset
+
+## Summary
+
+This dataset contains 14 training examples and 4
+held-out examples for **Clinical assistants need retrieval, source attribution, and explicit abstention before answers reach care teams.**
+
+Every record is synthetic and includes:
+
+- `input`: query, event, or feature description
+- `label`: expected class, route, relation, or evidence category
+- `context`: synthetic supporting context
+- `source`: fictional source identifier
+- `variant`: generation pattern
+- `synthetic`: always `true`
+
+## Uses
+
+- Reproducible unit and integration tests
+- Baseline model training
+- Evaluation harness development
+- Schema and architecture demonstrations
+
+## Limitations
+
+Synthetic educational data only. The baseline must not provide diagnosis, treatment, or emergency medical advice.
+
+This dataset does not represent real users, patients, customers, production
+traffic, or licensed media. It must not be presented as real-world evidence.
+
+## Related Model
+
+[RKB109/clinical-rag-safety-gateway-20260815-model](https://huggingface.co/RKB109/clinical-rag-safety-gateway-20260815-model)

data/test.jsonl

@@ -0,0 +1,4 @@
+{"id":"clinical-rag-safety-2-1","input":"What requires pharmacist review?","label":"medication-policy","context":"High-risk medication changes require pharmacist review before release.","source":"med-policy-02","variant":"direct","synthetic":true}
+{"id":"clinical-rag-safety-3-2","input":"In an operations review, When should urgent symptoms be escalated?","label":"triage-policy","context":"Urgent or life-threatening symptoms require immediate escalation to qualified clinical staff.","source":"triage-01","variant":"operations","synthetic":true}
+{"id":"clinical-rag-safety-4-3","input":"For an evaluation case, Can the assistant diagnose a patient?","label":"triage-policy","context":"The assistant must abstain from diagnosis and direct the user to a qualified clinician.","source":"triage-02","variant":"evaluation","synthetic":true}
+{"id":"clinical-rag-safety-6-1","input":"Can protected health data be used in test prompts?","label":"privacy-policy","context":"Use synthetic or properly de-identified records for testing and evaluation.","source":"privacy-02","variant":"direct","synthetic":true}

data/train.jsonl

@@ -0,0 +1,14 @@
+{"id":"clinical-rag-safety-1-1","input":"When should medication reconciliation be completed?","label":"medication-policy","context":"Complete medication reconciliation during admission, transfer, and discharge.","source":"med-policy-01","variant":"direct","synthetic":true}
+{"id":"clinical-rag-safety-1-2","input":"In an operations review, When should medication reconciliation be completed?","label":"medication-policy","context":"Complete medication reconciliation during admission, transfer, and discharge.","source":"med-policy-01","variant":"operations","synthetic":true}
+{"id":"clinical-rag-safety-1-3","input":"For an evaluation case, When should medication reconciliation be completed?","label":"medication-policy","context":"Complete medication reconciliation during admission, transfer, and discharge.","source":"med-policy-01","variant":"evaluation","synthetic":true}
+{"id":"clinical-rag-safety-2-2","input":"In an operations review, What requires pharmacist review?","label":"medication-policy","context":"High-risk medication changes require pharmacist review before release.","source":"med-policy-02","variant":"operations","synthetic":true}
+{"id":"clinical-rag-safety-2-3","input":"For an evaluation case, What requires pharmacist review?","label":"medication-policy","context":"High-risk medication changes require pharmacist review before release.","source":"med-policy-02","variant":"evaluation","synthetic":true}
+{"id":"clinical-rag-safety-3-1","input":"When should urgent symptoms be escalated?","label":"triage-policy","context":"Urgent or life-threatening symptoms require immediate escalation to qualified clinical staff.","source":"triage-01","variant":"direct","synthetic":true}
+{"id":"clinical-rag-safety-3-3","input":"For an evaluation case, When should urgent symptoms be escalated?","label":"triage-policy","context":"Urgent or life-threatening symptoms require immediate escalation to qualified clinical staff.","source":"triage-01","variant":"evaluation","synthetic":true}
+{"id":"clinical-rag-safety-4-1","input":"Can the assistant diagnose a patient?","label":"triage-policy","context":"The assistant must abstain from diagnosis and direct the user to a qualified clinician.","source":"triage-02","variant":"direct","synthetic":true}
+{"id":"clinical-rag-safety-4-2","input":"In an operations review, Can the assistant diagnose a patient?","label":"triage-policy","context":"The assistant must abstain from diagnosis and direct the user to a qualified clinician.","source":"triage-02","variant":"operations","synthetic":true}
+{"id":"clinical-rag-safety-5-1","input":"How should patient identifiers be handled?","label":"privacy-policy","context":"Use the minimum necessary patient information and never place identifiers in public logs.","source":"privacy-01","variant":"direct","synthetic":true}
+{"id":"clinical-rag-safety-5-2","input":"In an operations review, How should patient identifiers be handled?","label":"privacy-policy","context":"Use the minimum necessary patient information and never place identifiers in public logs.","source":"privacy-01","variant":"operations","synthetic":true}
+{"id":"clinical-rag-safety-5-3","input":"For an evaluation case, How should patient identifiers be handled?","label":"privacy-policy","context":"Use the minimum necessary patient information and never place identifiers in public logs.","source":"privacy-01","variant":"evaluation","synthetic":true}
+{"id":"clinical-rag-safety-6-2","input":"In an operations review, Can protected health data be used in test prompts?","label":"privacy-policy","context":"Use synthetic or properly de-identified records for testing and evaluation.","source":"privacy-02","variant":"operations","synthetic":true}
+{"id":"clinical-rag-safety-6-3","input":"For an evaluation case, Can protected health data be used in test prompts?","label":"privacy-policy","context":"Use synthetic or properly de-identified records for testing and evaluation.","source":"privacy-02","variant":"evaluation","synthetic":true}

project.json

@@ -0,0 +1,52 @@
+{
+  "name": "Clinical RAG Safety Gateway",
+  "problem": "Clinical assistants need retrieval, source attribution, and explicit abstention before answers reach care teams.",
+  "domain": "healthcare-ai",
+  "architecture": "retrieval",
+  "hugging_face_tasks": [
+    "question-answering",
+    "sentence-similarity",
+    "text-classification",
+    "summarization"
+  ],
+  "recommended_stack": [
+    "FastAPI for typed service endpoints",
+    "LangGraph for durable safety and escalation workflows",
+    "LlamaIndex for ingestion and retrieval abstractions",
+    "PostgreSQL plus pgvector for filtered vector search",
+    "Redis for caching and rate limiting",
+    "OpenTelemetry plus Phoenix for traces and evaluation"
+  ],
+  "real_world_data_sources": [
+    {
+      "name": "ClinicalTrials.gov API v2",
+      "url": "https://clinicaltrials.gov/api/v2/studies?pageSize=5",
+      "purpose": "Public study metadata for retrieval and citation tests"
+    },
+    {
+      "name": "openFDA drug labels",
+      "url": "https://api.fda.gov/drug/label.json?limit=5",
+      "purpose": "Public drug-label text for safety-oriented ingestion"
+    }
+  ],
+  "job_description_skills": [
+    "Production RAG design and retrieval evaluation",
+    "Healthcare AI safety, abstention, and auditability",
+    "Vector databases and metadata-aware retrieval",
+    "Agent orchestration with human review",
+    "LLMOps tracing, regression tests, and release gates"
+  ],
+  "impact_targets": [
+    "Reach Recall@5 >= 0.85 on a reviewed retrieval set",
+    "Maintain 100% citation coverage for non-abstained answers",
+    "Block 100% of diagnosis or emergency-advice test prompts",
+    "Keep p95 retrieval latency below 300 ms at 50 requests/second"
+  ],
+  "baseline_evaluation": {
+    "test_examples": 4,
+    "accuracy": 1,
+    "synthetic_evaluation": true
+  },
+  "estimated_delivery": "8-12 weeks for one engineer",
+  "generated_baseline_is_production_ready": false
+}

sources.json

@@ -0,0 +1,12 @@
+[
+  {
+    "name": "ClinicalTrials.gov API v2",
+    "url": "https://clinicaltrials.gov/api/v2/studies?pageSize=5",
+    "purpose": "Public study metadata for retrieval and citation tests"
+  },
+  {
+    "name": "openFDA drug labels",
+    "url": "https://api.fda.gov/drug/label.json?limit=5",
+    "purpose": "Public drug-label text for safety-oriented ingestion"
+  }
+]