Agent Patterns

import Anthropic from '@anthropic-ai/sdk';
import { OpenBio } from '@openbio/sdk';

const anthropic = new Anthropic();
const openbio = new OpenBio();

async function answerClinicalQuestion(subjectId: string, question: string) {
  // Step 1: Retrieve structured evidence context
  const context = await openbio.agents.evidenceContext({
    subjectId,
    format: 'structured',
    maxAssertions: 50,
    trustLayers: ['source_fact', 'normalized_fact'],
  });

  // Step 2: Build a grounded system prompt
  const systemPrompt = `You are a clinical evidence assistant. Answer questions using
only the evidence provided. For each claim, cite the assertion ID and trust layer.
Do not infer beyond what the evidence supports.

Evidence context:
${JSON.stringify(context, null, 2)}`;

  // Step 3: Call the LLM with the grounded context
  const response = await anthropic.messages.create({
    model: 'claude-opus-4-6',
    max_tokens: 1024,
    system: systemPrompt,
    messages: [{ role: 'user', content: question }],
  });

  return response.content[0].type === 'text' ? response.content[0].text : '';
}

// Usage:
const answer = await answerClinicalQuestion(
  'james-58',
  'What is this patient\'s hemoglobin trend over the past 6 months?',
);

import { OpenBio } from '@openbio/sdk';

const openbio = new OpenBio();

async function buildTrainingDataset() {
  // Find all subjects with anemia and an NSCLC-related diagnosis
  const cohort = await openbio.cohorts.query({
    criteria: [
      { type: 'lab_result', code: 'LOINC:718-7', value: { lt: 12 } },
      { type: 'diagnosis', code: 'SNOMED:254637007' },
    ],
    subjectTypes: ['human', 'canine', 'primate'],
    trustLayers: ['source_fact', 'normalized_fact'],
  });

  // Fetch full evidence for each matching subject
  const trainingData = await Promise.all(
    cohort.subjects.map(async (subject) => {
      const context = await openbio.agents.evidenceContext({
        subjectId: subject.id,
        format: 'structured',
        maxAssertions: 100,
        trustLayers: ['source_fact', 'normalized_fact'],
      });
      return {
        subjectId: subject.id,
        subjectType: subject.subjectType,
        assertions: context.assertions,
      };
    }),
  );

  return trainingData;
}

// Returns: array of { subjectId, subjectType, assertions[] }
// Ready for ingestion into a training pipeline

import { OpenBio } from '@openbio/sdk';

const openbio = new OpenBio();

async function formatCitation(assertionId: string): Promise<string> {
  const assertion = await openbio.evidence.get(assertionId);
  const chain = assertion.provenanceChain;

  // Build a structured citation string
  const source = chain[0]; // First node = raw source
  const collection = new Date(source.collectedAt).toLocaleDateString('en-US', {
    year: 'numeric',
    month: 'long',
    day: 'numeric',
  });

  return `[${assertion.display ?? assertion.type}: ${assertion.value} ${assertion.unit ?? ''} · ${source.source} · ${collection} · Trust: ${assertion.trustLayer} (${Math.round(assertion.confidence * 100)}%)]`;
}

// Example output:
// [Hemoglobin: 9.2 g/dL · Quest Diagnostics · March 15, 2025 · Trust: source_fact (99%)]

// Usage in agent response generation:
const citation = await formatCitation('assert-001');
const agentResponse = `The patient's hemoglobin is critically low at 9.2 g/dL ${citation}.`;