Smart Memoryfor Your AI

import vrin
from openai import OpenAI

# Initialize Vrin Memory Orchestrator
vrin_client = vrin.Client(api_key="your-api-key")

# Doctor records new patient episode
episode_data = {
  "patient_id": "patient_789",
  "conversation": "Patient reports worsening chest pain, family history of heart disease...",
  "timestamp": "2024-01-15T14:30:00Z",
  "provider": "Dr. Smith"
}

# 1. Doctor hits submit -> Vrin API called
response = vrin_client.episodes.create(
  data=episode_data,
  extract_facts=True,
  build_relationships=True
)

# 2. Vrin extracts facts & causal relationships (memory-efficient)
extracted_facts = response.facts
# Example: ["Patient: chest pain worsening", "Family history: heart disease", 
#          "Relationship: genetic_risk_factor"]

# 3. Store only essential facts in vector DB (90% storage reduction)
vrin_client.memory.store_facts(
  patient_id="patient_789",
  facts=extracted_facts,
  compress=True  # Memory-efficient storage
)

# 4. Later: Doctor needs patient info
query = "Show me this patient's cardiac risk factors and recent symptoms"

# 5. Retrieve relevant facts based on search query
relevant_facts = vrin_client.memory.search(
  patient_id="patient_789",
  query=query,
  max_results=20
)

# 6. Create knowledge graph from retrieved facts
knowledge_graph = vrin_client.graph.build(
  facts=relevant_facts,
  include_relationships=True
)

# 7. LLM summarizes with RL optimization & bandit prompt selection
summary = vrin_client.insights.generate(
  knowledge_graph=knowledge_graph,
  query=query,
  format="clinical_summary",
  optimize_prompt=True,  # RL-driven prompt selection
  bandit_optimization=True  # Continuous learning
)

print(summary.content)
# Output: "Patient has elevated cardiac risk: family history of CAD, 
# current chest pain symptoms increasing in frequency..."