Phase 1 & 2: Authentication and Children Management

Completed Features: - Full JWT authentication system with refresh tokens - User registration and login with device fingerprinting - Child profile CRUD operations with permission-based access - Family management with roles and permissions - Database migrations for core auth and family structure - Comprehensive test coverage (37 unit + E2E tests) Tech Stack: - NestJS backend with TypeORM - PostgreSQL database - JWT authentication with Passport - bcrypt password hashing - Docker Compose for infrastructure 🤖 Generated with Claude Code
2025-09-30 18:40:10 +03:00
commit 98e01ebe80
35 changed files with 9683 additions and 0 deletions
--- a/docs/maternal-app-ai-context.md
+++ b/docs/maternal-app-ai-context.md
@@ -0,0 +1,551 @@
+# AI Context & Prompting Templates - Maternal Organization App
+
+## LangChain Configuration
+
+### Core Setup
+```typescript
+// services/ai/langchainConfig.ts
+import { ChatOpenAI } from 'langchain/chat_models/openai';
+import { ConversationSummaryMemory } from 'langchain/memory';
+import { PromptTemplate } from 'langchain/prompts';
+import { LLMChain } from 'langchain/chains';
+
+export const initializeLangChain = () => {
+  const model = new ChatOpenAI({
+    modelName: 'gpt-4-turbo-preview',
+    temperature: 0.7,
+    maxTokens: 1000,
+    openAIApiKey: process.env.OPENAI_API_KEY,
+    callbacks: [
+      {
+        handleLLMStart: async (llm, prompts) => {
+          logger.info('LLM request started', { promptLength: prompts[0].length });
+        },
+        handleLLMEnd: async (output) => {
+          logger.info('LLM request completed', { tokensUsed: output.llmOutput?.tokenUsage });
+        },
+      },
+    ],
+  });
+
+  const memory = new ConversationSummaryMemory({
+    memoryKey: 'chat_history',
+    llm: model,
+    maxTokenLimit: 2000,
+  });
+
+  return { model, memory };
+};
+```
+
+### Context Window Management
+```typescript
+class ContextManager {
+  private maxContextTokens = 4000;
+  private priorityWeights = {
+    currentQuery: 1.0,
+    recentActivities: 0.8,
+    childProfile: 0.7,
+    historicalPatterns: 0.6,
+    generalGuidelines: 0.4,
+  };
+
+  async buildContext(
+    query: string,
+    childId: string,
+    userId: string
+  ): Promise<AIContext> {
+    const contexts = await Promise.all([
+      this.getChildProfile(childId),
+      this.getRecentActivities(childId, 48), // Last 48 hours
+      this.getPatterns(childId),
+      this.getParentPreferences(userId),
+      this.getPreviousConversation(userId, childId),
+    ]);
+
+    return this.prioritizeContext(contexts, query);
+  }
+
+  private prioritizeContext(
+    contexts: ContextPart[],
+    query: string
+  ): AIContext {
+    // Token counting
+    let currentTokens = this.countTokens(query);
+    const prioritizedContexts: ContextPart[] = [];
+
+    // Sort by relevance and priority
+    const sorted = contexts
+      .map(ctx => ({
+        ...ctx,
+        score: this.calculateRelevance(ctx, query) * this.priorityWeights[ctx.type],
+      }))
+      .sort((a, b) => b.score - a.score);
+
+    // Add contexts until token limit
+    for (const context of sorted) {
+      const tokens = this.countTokens(context.content);
+      if (currentTokens + tokens <= this.maxContextTokens) {
+        prioritizedContexts.push(context);
+        currentTokens += tokens;
+      }
+    }
+
+    return {
+      query,
+      contexts: prioritizedContexts,
+      totalTokens: currentTokens,
+    };
+  }
+}
+```
+
+---
+
+## System Prompts
+
+### Base System Prompt
+```typescript
+const BASE_SYSTEM_PROMPT = `You are a knowledgeable, empathetic AI assistant helping parents care for their children aged 0-6 years. 
+
+Core Guidelines:
+1. SAFETY FIRST: Never provide medical diagnoses. Always recommend consulting healthcare providers for medical concerns.
+2. EVIDENCE-BASED: Provide advice based on pediatric best practices and research.
+3. PARENT-SUPPORTIVE: Be encouraging and non-judgmental. Every family is different.
+4. PRACTICAL: Give actionable, realistic suggestions that work for busy parents.
+5. CULTURALLY AWARE: Respect diverse parenting approaches and cultural practices.
+
+You have access to:
+- The child's recent activity data (feeding, sleep, diapers)
+- Developmental milestones for their age
+- Pattern analysis from their historical data
+- Family preferences and routines
+
+Response Guidelines:
+- Keep responses concise (under 150 words unless asked for detail)
+- Use simple, clear language (avoid medical jargon)
+- Provide specific, actionable suggestions
+- Acknowledge parent concerns with empathy
+- Include relevant safety warnings when appropriate`;
+```
+
+### Child-Specific Context Template
+```typescript
+const CHILD_CONTEXT_TEMPLATE = `Child Profile:
+- Name: {childName}
+- Age: {ageInMonths} months ({ageInYears} years)
+- Developmental Stage: {developmentalStage}
+- Known Conditions: {medicalConditions}
+- Allergies: {allergies}
+
+Recent Patterns (last 7 days):
+- Average sleep: {avgSleepHours} hours/day
+- Feeding frequency: Every {feedingInterval} hours
+- Growth trajectory: {growthPercentile} percentile
+
+Current Concerns:
+{parentConcerns}
+
+Recent Activities (last 24 hours):
+{recentActivities}`;
+```
+
+---
+
+## Prompt Templates by Scenario
+
+### Sleep-Related Queries
+```typescript
+const SLEEP_PROMPT = PromptTemplate.fromTemplate(`
+${BASE_SYSTEM_PROMPT}
+
+Context:
+{childContext}
+
+Sleep-Specific Data:
+- Recent bedtimes: {recentBedtimes}
+- Wake windows today: {wakeWindows}
+- Nap schedule: {napSchedule}
+- Sleep regression risk: {regressionRisk}
+
+Parent Question: {question}
+
+Provide practical sleep advice considering:
+1. Age-appropriate wake windows
+2. Recent sleep patterns
+3. Common sleep regressions at this age
+4. Environmental factors
+
+Response:
+`);
+```
+
+### Feeding Queries
+```typescript
+const FEEDING_PROMPT = PromptTemplate.fromTemplate(`
+${BASE_SYSTEM_PROMPT}
+
+Context:
+{childContext}
+
+Feeding-Specific Data:
+- Feeding type: {feedingType}
+- Recent intake: {recentIntake}
+- Growth trend: {growthTrend}
+- Solid foods status: {solidsStatus}
+
+Parent Question: {question}
+
+Provide feeding guidance considering:
+1. Age-appropriate feeding amounts
+2. Growth patterns
+3. Feeding milestones
+4. Any mentioned concerns
+
+Response:
+`);
+```
+
+### Developmental Milestones
+```typescript
+const MILESTONE_PROMPT = PromptTemplate.fromTemplate(`
+${BASE_SYSTEM_PROMPT}
+
+Context:
+{childContext}
+
+Developmental Data:
+- Expected milestones: {expectedMilestones}
+- Achieved milestones: {achievedMilestones}
+- Areas of focus: {developmentalFocus}
+
+Parent Question: {question}
+
+Provide developmental guidance:
+1. What's typical for this age
+2. Activities to encourage development
+3. When to consult professionals
+4. Celebrate achievements while avoiding comparison
+
+Response:
+`);
+```
+
+### Behavioral Concerns
+```typescript
+const BEHAVIOR_PROMPT = PromptTemplate.fromTemplate(`
+${BASE_SYSTEM_PROMPT}
+
+Context:
+{childContext}
+
+Behavioral Patterns:
+- Recent behaviors: {recentBehaviors}
+- Triggers identified: {triggers}
+- Sleep/hunger status: {physiologicalState}
+
+Parent Question: {question}
+
+Provide behavioral guidance:
+1. Age-appropriate expectations
+2. Positive parenting strategies
+3. Understanding underlying needs
+4. Consistency and routine importance
+
+Response:
+`);
+```
+
+---
+
+## Safety Boundaries
+
+### Medical Disclaimer Triggers
+```typescript
+const MEDICAL_TRIGGERS = [
+  'diagnose', 'disease', 'syndrome', 'disorder',
+  'medication', 'dosage', 'prescription',
+  'emergency', 'urgent', 'bleeding', 'unconscious',
+  'seizure', 'fever over 104', 'difficulty breathing',
+];
+
+const MEDICAL_DISCLAIMER = `I understand you're concerned about {concern}. This seems like a medical issue that requires professional evaluation. Please contact your pediatrician or healthcare provider immediately. If this is an emergency, call your local emergency number.
+
+For reference, here are signs requiring immediate medical attention:
+- Difficulty breathing or bluish skin
+- Unresponsiveness or difficulty waking
+- High fever (over 104°F/40°C)
+- Severe dehydration
+- Head injury with vomiting or confusion`;
+```
+
+### Mental Health Support
+```typescript
+const MENTAL_HEALTH_PROMPT = `I hear that you're going through a difficult time. Your feelings are valid and important. 
+
+Here are some immediate resources:
+- Postpartum Support International: 1-800-4-PPD-MOMS
+- Crisis Text Line: Text HOME to 741741
+- Local support groups: {localResources}
+
+Please consider reaching out to:
+- Your healthcare provider
+- A mental health professional
+- Trusted family or friends
+
+Remember: Seeking help is a sign of strength, not weakness. Your wellbeing matters for both you and your baby.`;
+```
+
+---
+
+## Response Formatting
+
+### Structured Response Template
+```typescript
+interface AIResponse {
+  mainAnswer: string;
+  keyPoints?: string[];
+  actionItems?: string[];
+  safetyNotes?: string[];
+  relatedResources?: Resource[];
+  confidenceLevel: 'high' | 'medium' | 'low';
+  shouldEscalate: boolean;
+}
+
+const formatResponse = (raw: string, metadata: ResponseMetadata): AIResponse => {
+  return {
+    mainAnswer: extractMainAnswer(raw),
+    keyPoints: extractBulletPoints(raw),
+    actionItems: extractActionItems(raw),
+    safetyNotes: extractSafetyWarnings(raw),
+    relatedResources: findRelevantResources(metadata),
+    confidenceLevel: calculateConfidence(metadata),
+    shouldEscalate: checkEscalationTriggers(raw),
+  };
+};
+```
+
+### Localized Response Generation
+```typescript
+const LOCALIZED_PROMPTS = {
+  'en-US': {
+    greeting: "I understand your concern about {topic}.",
+    transition: "Based on {childName}'s patterns,",
+    closing: "Remember, every baby is different.",
+  },
+  'es-ES': {
+    greeting: "Entiendo tu preocupación sobre {topic}.",
+    transition: "Basándome en los patrones de {childName},",
+    closing: "Recuerda, cada bebé es diferente.",
+  },
+  'fr-FR': {
+    greeting: "Je comprends votre inquiétude concernant {topic}.",
+    transition: "D'après les habitudes de {childName},",
+    closing: "Rappelez-vous, chaque bébé est unique.",
+  },
+};
+```
+
+---
+
+## Personalization Engine
+
+### Learning from Feedback
+```typescript
+class PersonalizationEngine {
+  async updateResponsePreferences(
+    userId: string,
+    feedback: UserFeedback
+  ) {
+    const preferences = await this.getUserPreferences(userId);
+    
+    // Update preference weights
+    if (feedback.helpful) {
+      preferences.preferredResponseLength = feedback.responseLength;
+      preferences.preferredDetailLevel = feedback.detailLevel;
+      preferences.preferredTone = feedback.tone;
+    }
+    
+    // Learn from negative feedback
+    if (!feedback.helpful && feedback.reason) {
+      this.adjustPromptTemplate(preferences, feedback.reason);
+    }
+    
+    await this.saveUserPreferences(userId, preferences);
+  }
+
+  private adjustPromptTemplate(
+    preferences: UserPreferences,
+    reason: string
+  ): PromptTemplate {
+    const adjustments = {
+      'too_technical': { jargonLevel: 'minimal', explanationStyle: 'simple' },
+      'too_general': { specificityLevel: 'high', includeExamples: true },
+      'too_long': { maxLength: 100, bulletPoints: true },
+      'not_actionable': { focusOnActions: true, includeSteps: true },
+    };
+    
+    return this.applyAdjustments(preferences, adjustments[reason]);
+  }
+}
+```
+
+---
+
+## Chain of Thought for Complex Queries
+
+### Multi-Step Reasoning
+```typescript
+const COMPLEX_REASONING_PROMPT = `Let me analyze this step-by-step:
+
+Step 1: Understanding the Situation
+{situationAnalysis}
+
+Step 2: Identifying Patterns
+Looking at {childName}'s recent data:
+{patternAnalysis}
+
+Step 3: Considering Age-Appropriate Norms
+For a {ageInMonths}-month-old:
+{developmentalNorms}
+
+Step 4: Practical Recommendations
+Based on the above:
+{recommendations}
+
+Step 5: What to Monitor
+Keep track of:
+{monitoringPoints}`;
+```
+
+---
+
+## Conversation Memory Management
+
+### Memory Summarization
+```typescript
+class ConversationMemory {
+  private maxConversationLength = 10;
+  
+  async summarizeConversation(
+    messages: Message[],
+    childId: string
+  ): Promise<string> {
+    if (messages.length <= 3) {
+      return messages.map(m => m.content).join('\n');
+    }
+    
+    const summary = await this.llm.summarize({
+      messages,
+      focusPoints: [
+        'Main concerns discussed',
+        'Advice given',
+        'Action items suggested',
+        'Follow-up needed',
+      ],
+    });
+    
+    return summary;
+  }
+  
+  async getRelevantHistory(
+    userId: string,
+    childId: string,
+    currentQuery: string
+  ): Promise<string> {
+    const history = await this.fetchHistory(userId, childId);
+    
+    // Semantic search for relevant past conversations
+    const relevant = await this.semanticSearch(history, currentQuery);
+    
+    return this.formatHistory(relevant);
+  }
+}
+```
+
+---
+
+## Prompt Injection Protection
+
+### Security Filters
+```typescript
+const INJECTION_PATTERNS = [
+  /ignore previous instructions/i,
+  /system:/i,
+  /admin mode/i,
+  /bypass safety/i,
+  /pretend you are/i,
+];
+
+const sanitizeUserInput = (input: string): string => {
+  // Check for injection attempts
+  for (const pattern of INJECTION_PATTERNS) {
+    if (pattern.test(input)) {
+      logger.warn('Potential prompt injection detected', { input });
+      return 'Please ask a question about childcare.';
+    }
+  }
+  
+  // Escape special characters
+  return input
+    .replace(/[<>]/g, '')
+    .substring(0, 500); // Limit length
+};
+```
+
+---
+
+## Testing Prompt Effectiveness
+
+### Prompt Evaluation Metrics
+```typescript
+interface PromptMetrics {
+  relevance: number;      // 0-1: Response answers the question
+  safety: number;         // 0-1: Appropriate medical disclaimers
+  actionability: number;  // 0-1: Practical suggestions provided
+  empathy: number;        // 0-1: Supportive tone
+  accuracy: number;       // 0-1: Factually correct
+}
+
+const evaluatePromptResponse = async (
+  prompt: string,
+  response: string,
+  expectedQualities: PromptMetrics
+): Promise<EvaluationResult> => {
+  const evaluation = await this.evaluatorLLM.evaluate({
+    prompt,
+    response,
+    criteria: expectedQualities,
+  });
+  
+  return {
+    passed: evaluation.overall > 0.8,
+    metrics: evaluation,
+    suggestions: evaluation.improvements,
+  };
+};
+```
+
+### Test Cases
+```typescript
+const promptTestCases = [
+  {
+    scenario: 'Sleep regression concern',
+    input: 'My 4-month-old suddenly won\'t sleep',
+    expectedResponse: {
+      containsMention: ['4-month sleep regression', 'normal', 'temporary'],
+      includesActions: ['consistent bedtime', 'wake windows'],
+      avoidsMedical: true,
+    },
+  },
+  {
+    scenario: 'Feeding amount worry',
+    input: 'Is 4oz enough for my 2-month-old?',
+    expectedResponse: {
+      containsMention: ['varies by baby', 'weight gain', 'wet diapers'],
+      includesActions: ['track intake', 'consult pediatrician'],
+      providesRanges: true,
+    },
+  },
+];
+```