@neuroequality/neuroadapt-ai/dist/index.cjs.map

AI-powered accessibility personalization for neurodivergent users

{"version":3,"file":"index.cjs","sources":["../src/adapters/base-adapter.ts","../src/adapters/openai-adapter.ts","../src/adapters/claude-adapter.ts","../src/adapters/ollama-adapter.ts","../src/predictable/predictable-ai.ts","../src/advanced/adaptive-engine.ts","../src/advanced/neural-adaptation.ts","../src/advanced/real-time-optimization.ts","../src/enterprise/sso-integration.ts","../src/enterprise/analytics-dashboard.ts","../src/enterprise/deployment-tools.ts","../src/index.ts"],"sourcesContent":["import type { AIAdapter, AIAdapterConfig, AIMessage, AIResponse, AICompletionOptions, AIStreamChunk } from '../types/index.js';\n\nexport abstract class BaseAIAdapter implements AIAdapter {\n  protected config: AIAdapterConfig;\n  protected currentModel: string;\n\n  constructor(config: AIAdapterConfig) {\n    this.config = config;\n    this.currentModel = config.model || this.getDefaultModel();\n  }\n\n  abstract get name(): string;\n  abstract get models(): string[];\n  protected abstract getDefaultModel(): string;\n\n  abstract complete(messages: AIMessage[], options?: AICompletionOptions): Promise<AIResponse>;\n  abstract stream?(messages: AIMessage[], options?: AICompletionOptions): AsyncIterable<AIStreamChunk>;\n  abstract isAvailable(): Promise<boolean>;\n\n  getModel(): string {\n    return this.currentModel;\n  }\n\n  setModel(model: string): void {\n    if (!this.models.includes(model)) {\n      throw new Error(`Model ${model} is not supported by ${this.name}`);\n    }\n    this.currentModel = model;\n  }\n\n  protected async retry<T>(\n    operation: () => Promise<T>,\n    attempts = this.config.retryAttempts || 3,\n    delay = this.config.retryDelay || 1000\n  ): Promise<T> {\n    let lastError: Error;\n\n    for (let attempt = 1; attempt <= attempts; attempt++) {\n      try {\n        return await operation();\n      } catch (error) {\n        lastError = error as Error;\n        \n        if (attempt === attempts) {\n          throw lastError;\n        }\n\n        if (this.isRetryableError(error)) {\n          await this.sleep(delay * Math.pow(2, attempt - 1));\n        } else {\n          throw lastError;\n        }\n      }\n    }\n\n    throw lastError!;\n  }\n\n  protected isRetryableError(error: unknown): boolean {\n    if (error instanceof Error) {\n      const message = error.message.toLowerCase();\n      return (\n        message.includes('timeout') ||\n        message.includes('rate limit') ||\n        message.includes('503') ||\n        message.includes('502') ||\n        message.includes('500')\n      );\n    }\n    return false;\n  }\n\n  protected sleep(ms: number): Promise<void> {\n    return new Promise(resolve => setTimeout(resolve, ms));\n  }\n\n  protected formatMessages(messages: AIMessage[]): unknown[] {\n    return messages.map(msg => ({\n      role: msg.role,\n      content: msg.content,\n    }));\n  }\n\n  protected createAbortController(signal?: AbortSignal, timeout?: number): AbortController {\n    const controller = new AbortController();\n    \n    if (signal) {\n      signal.addEventListener('abort', () => controller.abort());\n    }\n    \n    if (timeout) {\n      setTimeout(() => controller.abort(), timeout);\n    }\n    \n    return controller;\n  }\n}","import { BaseAIAdapter } from './base-adapter.js';\nimport type { AIMessage, AIResponse, AICompletionOptions, AIStreamChunk, AIAdapterConfig } from '../types/index.js';\n\nexport interface OpenAIAdapterConfig extends AIAdapterConfig {\n  organization?: string;\n}\n\nexport class OpenAIAdapter extends BaseAIAdapter {\n  private openai: any; // Will be dynamically imported\n\n  constructor(config: OpenAIAdapterConfig) {\n    super(config);\n    this.initializeOpenAI();\n  }\n\n  get name(): string {\n    return 'OpenAI';\n  }\n\n  get models(): string[] {\n    return [\n      'gpt-4',\n      'gpt-4-turbo',\n      'gpt-4-turbo-preview',\n      'gpt-3.5-turbo',\n      'gpt-3.5-turbo-16k',\n    ];\n  }\n\n  protected getDefaultModel(): string {\n    return 'gpt-4-turbo-preview';\n  }\n\n  async complete(messages: AIMessage[], options: AICompletionOptions = {}): Promise<AIResponse> {\n    if (!this.openai) {\n      throw new Error('OpenAI client not initialized. Make sure to install the openai package.');\n    }\n\n    return this.retry(async () => {\n      const controller = this.createAbortController(options.abortSignal, this.config.timeout);\n\n      const response = await this.openai.chat.completions.create({\n        model: this.currentModel,\n        messages: this.formatMessages(messages),\n        max_tokens: options.maxTokens,\n        temperature: options.temperature,\n        top_p: options.topP,\n        stop: options.stop,\n        stream: false,\n      }, {\n        signal: controller.signal,\n      });\n\n      return {\n        content: response.choices[0]?.message?.content || '',\n        usage: {\n          promptTokens: response.usage?.prompt_tokens || 0,\n          completionTokens: response.usage?.completion_tokens || 0,\n          totalTokens: response.usage?.total_tokens || 0,\n        },\n        finishReason: response.choices[0]?.finish_reason as AIResponse['finishReason'],\n        metadata: {\n          model: response.model,\n          created: response.created,\n        },\n      };\n    });\n  }\n\n  async *stream(messages: AIMessage[], options: AICompletionOptions = {}): AsyncIterable<AIStreamChunk> {\n    if (!this.openai) {\n      throw new Error('OpenAI client not initialized. Make sure to install the openai package.');\n    }\n\n    const controller = this.createAbortController(options.abortSignal, this.config.timeout);\n\n    const stream = await this.openai.chat.completions.create({\n      model: this.currentModel,\n      messages: this.formatMessages(messages),\n      max_tokens: options.maxTokens,\n      temperature: options.temperature,\n      top_p: options.topP,\n      stop: options.stop,\n      stream: true,\n    }, {\n      signal: controller.signal,\n    });\n\n    for await (const chunk of stream) {\n      const delta = chunk.choices[0]?.delta?.content || '';\n      const done = chunk.choices[0]?.finish_reason !== null;\n\n      yield {\n        delta,\n        done,\n        usage: chunk.usage ? {\n          promptTokens: chunk.usage.prompt_tokens || 0,\n          completionTokens: chunk.usage.completion_tokens || 0,\n          totalTokens: chunk.usage.total_tokens || 0,\n        } : undefined,\n      };\n\n      if (done) break;\n    }\n  }\n\n  async isAvailable(): Promise<boolean> {\n    try {\n      if (!this.openai) {\n        return false;\n      }\n      await this.openai.models.list();\n      return true;\n    } catch (error) {\n      return false;\n    }\n  }\n\n  private async initializeOpenAI(): Promise<void> {\n    try {\n      const { default: OpenAI } = await import('openai');\n      this.openai = new OpenAI({\n        apiKey: this.config.apiKey,\n        baseURL: this.config.baseURL,\n        organization: (this.config as OpenAIAdapterConfig).organization,\n      });\n    } catch (error) {\n      console.warn('OpenAI package not found. Install it to use OpenAI adapter.');\n    }\n  }\n}","import { BaseAIAdapter } from './base-adapter.js';\nimport type { AIMessage, AIResponse, AICompletionOptions, AIStreamChunk, AIAdapterConfig } from '../types/index.js';\n\nexport class ClaudeAdapter extends BaseAIAdapter {\n  private anthropic: any; // Will be dynamically imported\n\n  constructor(config: AIAdapterConfig) {\n    super(config);\n    this.initializeAnthropic();\n  }\n\n  get name(): string {\n    return 'Claude';\n  }\n\n  get models(): string[] {\n    return [\n      'claude-3-5-sonnet-20241022',\n      'claude-3-5-haiku-20241022',\n      'claude-3-opus-20240229',\n      'claude-3-sonnet-20240229',\n      'claude-3-haiku-20240307',\n    ];\n  }\n\n  protected getDefaultModel(): string {\n    return 'claude-3-5-sonnet-20241022';\n  }\n\n  async complete(messages: AIMessage[], options: AICompletionOptions = {}): Promise<AIResponse> {\n    if (!this.anthropic) {\n      throw new Error('Anthropic SDK not initialized. Make sure to install @anthropic-ai/sdk package.');\n    }\n\n    return this.retry(async () => {\n      const { system, messages: formattedMessages } = this.formatClaudeMessages(messages);\n\n      const response = await this.anthropic.messages.create({\n        model: this.currentModel,\n        max_tokens: options.maxTokens || 4096,\n        temperature: options.temperature,\n        top_p: options.topP,\n        stop_sequences: options.stop,\n        system,\n        messages: formattedMessages,\n        stream: false,\n      });\n\n      return {\n        content: response.content[0]?.text || '',\n        usage: {\n          promptTokens: response.usage?.input_tokens || 0,\n          completionTokens: response.usage?.output_tokens || 0,\n          totalTokens: (response.usage?.input_tokens || 0) + (response.usage?.output_tokens || 0),\n        },\n        finishReason: response.stop_reason as AIResponse['finishReason'],\n        metadata: {\n          model: response.model,\n          role: response.role,\n        },\n      };\n    });\n  }\n\n  async *stream(messages: AIMessage[], options: AICompletionOptions = {}): AsyncIterable<AIStreamChunk> {\n    if (!this.anthropic) {\n      throw new Error('Anthropic SDK not initialized. Make sure to install @anthropic-ai/sdk package.');\n    }\n\n    const { system, messages: formattedMessages } = this.formatClaudeMessages(messages);\n\n    const stream = await this.anthropic.messages.create({\n      model: this.currentModel,\n      max_tokens: options.maxTokens || 4096,\n      temperature: options.temperature,\n      top_p: options.topP,\n      stop_sequences: options.stop,\n      system,\n      messages: formattedMessages,\n      stream: true,\n    });\n\n    for await (const event of stream) {\n      if (event.type === 'content_block_delta') {\n        yield {\n          delta: event.delta?.text || '',\n          done: false,\n        };\n      } else if (event.type === 'message_stop') {\n        yield {\n          delta: '',\n          done: true,\n          usage: event.message?.usage ? {\n            promptTokens: event.message.usage.input_tokens || 0,\n            completionTokens: event.message.usage.output_tokens || 0,\n            totalTokens: (event.message.usage.input_tokens || 0) + (event.message.usage.output_tokens || 0),\n          } : undefined,\n        };\n        break;\n      }\n    }\n  }\n\n  async isAvailable(): Promise<boolean> {\n    try {\n      if (!this.anthropic) {\n        return false;\n      }\n      // Simple test to check if the API key is valid\n      await this.anthropic.messages.create({\n        model: this.currentModel,\n        max_tokens: 1,\n        messages: [{ role: 'user', content: 'test' }],\n      });\n      return true;\n    } catch (error) {\n      // If it's an auth error but the SDK is working, return false\n      // If it's a network error, also return false\n      return false;\n    }\n  }\n\n  private formatClaudeMessages(messages: AIMessage[]): { system?: string; messages: Array<{ role: string; content: string }> } {\n    let system: string | undefined;\n    const formattedMessages: Array<{ role: string; content: string }> = [];\n\n    for (const message of messages) {\n      if (message.role === 'system') {\n        system = message.content;\n      } else {\n        formattedMessages.push({\n          role: message.role,\n          content: message.content,\n        });\n      }\n    }\n\n    return { system, messages: formattedMessages };\n  }\n\n  private async initializeAnthropic(): Promise<void> {\n    try {\n      const { default: Anthropic } = await import('@anthropic-ai/sdk');\n      this.anthropic = new Anthropic({\n        apiKey: this.config.apiKey,\n        baseURL: this.config.baseURL,\n      });\n    } catch (error) {\n      console.warn('Anthropic SDK not found. Install @anthropic-ai/sdk to use Claude adapter.');\n    }\n  }\n}","import { BaseAIAdapter } from './base-adapter.js';\nimport type { AIMessage, AIResponse, AICompletionOptions, AIStreamChunk, AIAdapterConfig } from '../types/index.js';\n\nexport interface OllamaAdapterConfig extends AIAdapterConfig {\n  baseURL?: string;\n}\n\nexport class OllamaAdapter extends BaseAIAdapter {\n  private baseURL: string;\n\n  constructor(config: OllamaAdapterConfig) {\n    super(config);\n    this.baseURL = config.baseURL || 'http://localhost:11434';\n  }\n\n  get name(): string {\n    return 'Ollama';\n  }\n\n  get models(): string[] {\n    return [\n      'deepseek-r1:32b',\n      'llama3.2:3b',\n      'llama3.2:1b',\n      'phi3:mini',\n      'mistral:7b',\n      'codellama:7b',\n      'gemma2:2b',\n    ];\n  }\n\n  protected getDefaultModel(): string {\n    return 'deepseek-r1:32b';\n  }\n\n  async complete(messages: AIMessage[], options: AICompletionOptions = {}): Promise<AIResponse> {\n    return this.retry(async () => {\n      const controller = this.createAbortController(options.abortSignal, this.config.timeout);\n\n      const response = await fetch(`${this.baseURL}/api/chat`, {\n        method: 'POST',\n        headers: {\n          'Content-Type': 'application/json',\n        },\n        body: JSON.stringify({\n          model: this.currentModel,\n          messages: this.formatMessages(messages),\n          stream: false,\n          options: {\n            temperature: options.temperature,\n            top_p: options.topP,\n            stop: options.stop,\n            num_predict: options.maxTokens,\n          },\n        }),\n        signal: controller.signal,\n      });\n\n      if (!response.ok) {\n        throw new Error(`Ollama API error: ${response.status} ${response.statusText}`);\n      }\n\n      const data = await response.json();\n\n      return {\n        content: data.message?.content || '',\n        usage: {\n          promptTokens: data.prompt_eval_count || 0,\n          completionTokens: data.eval_count || 0,\n          totalTokens: (data.prompt_eval_count || 0) + (data.eval_count || 0),\n        },\n        finishReason: data.done ? 'stop' : 'length',\n        metadata: {\n          model: data.model,\n          created_at: data.created_at,\n          total_duration: data.total_duration,\n          load_duration: data.load_duration,\n          prompt_eval_duration: data.prompt_eval_duration,\n          eval_duration: data.eval_duration,\n        },\n      };\n    });\n  }\n\n  async *stream(messages: AIMessage[], options: AICompletionOptions = {}): AsyncIterable<AIStreamChunk> {\n    const controller = this.createAbortController(options.abortSignal, this.config.timeout);\n\n    const response = await fetch(`${this.baseURL}/api/chat`, {\n      method: 'POST',\n      headers: {\n        'Content-Type': 'application/json',\n      },\n      body: JSON.stringify({\n        model: this.currentModel,\n        messages: this.formatMessages(messages),\n        stream: true,\n        options: {\n          temperature: options.temperature,\n          top_p: options.topP,\n          stop: options.stop,\n          num_predict: options.maxTokens,\n        },\n      }),\n      signal: controller.signal,\n    });\n\n    if (!response.ok) {\n      throw new Error(`Ollama API error: ${response.status} ${response.statusText}`);\n    }\n\n    const reader = response.body?.getReader();\n    if (!reader) {\n      throw new Error('No response body');\n    }\n\n    const decoder = new TextDecoder();\n    let buffer = '';\n\n    try {\n      while (true) {\n        const { done, value } = await reader.read();\n        if (done) break;\n\n        buffer += decoder.decode(value, { stream: true });\n        const lines = buffer.split('\\n');\n        buffer = lines.pop() || '';\n\n        for (const line of lines) {\n          if (line.trim()) {\n            try {\n              const chunk = JSON.parse(line);\n              const delta = chunk.message?.content || '';\n              const isDone = chunk.done || false;\n\n              yield {\n                delta,\n                done: isDone,\n                usage: isDone ? {\n                  promptTokens: chunk.prompt_eval_count || 0,\n                  completionTokens: chunk.eval_count || 0,\n                  totalTokens: (chunk.prompt_eval_count || 0) + (chunk.eval_count || 0),\n                } : undefined,\n              };\n\n              if (isDone) return;\n            } catch (parseError) {\n              console.warn('Failed to parse Ollama chunk:', line);\n            }\n          }\n        }\n      }\n    } finally {\n      reader.releaseLock();\n    }\n  }\n\n  async isAvailable(): Promise<boolean> {\n    try {\n      const response = await fetch(`${this.baseURL}/api/tags`, {\n        method: 'GET',\n        signal: AbortSignal.timeout(5000),\n      });\n      return response.ok;\n    } catch (error) {\n      return false;\n    }\n  }\n\n  async getAvailableModels(): Promise<string[]> {\n    try {\n      const response = await fetch(`${this.baseURL}/api/tags`);\n      if (!response.ok) {\n        throw new Error('Failed to fetch models');\n      }\n\n      const data = await response.json();\n      return data.models?.map((model: { name: string }) => model.name) || [];\n    } catch (error) {\n      console.warn('Failed to fetch available models from Ollama:', error);\n      return this.models;\n    }\n  }\n\n  async pullModel(model: string): Promise<void> {\n    const response = await fetch(`${this.baseURL}/api/pull`, {\n      method: 'POST',\n      headers: {\n        'Content-Type': 'application/json',\n      },\n      body: JSON.stringify({ name: model }),\n    });\n\n    if (!response.ok) {\n      throw new Error(`Failed to pull model ${model}: ${response.statusText}`);\n    }\n\n    // Stream the pull progress\n    const reader = response.body?.getReader();\n    if (!reader) return;\n\n    const decoder = new TextDecoder();\n    try {\n      while (true) {\n        const { done, value } = await reader.read();\n        if (done) break;\n\n        const chunk = decoder.decode(value);\n        console.log('Pull progress:', chunk);\n      }\n    } finally {\n      reader.releaseLock();\n    }\n  }\n}","import { EventEmitter } from 'eventemitter3';\nimport type { CognitivePreferences } from '@neuroadapt/core';\nimport type {\n  AIAdapter,\n  AIMessage,\n  AIResponse,\n  AIStreamChunk,\n  PredictableAIConfig,\n  PredictableAIEvents,\n  UndoState,\n  AICompletionOptions,\n} from '../types/index.js';\n\nexport class PredictableAI extends EventEmitter<PredictableAIEvents> {\n  private adapter: AIAdapter;\n  private config: PredictableAIConfig;\n  private undoHistory: UndoState[] = [];\n  private currentStep = 0;\n  private cache = new Map<string, AIResponse>();\n\n  constructor(adapter: AIAdapter, config: Partial<PredictableAIConfig> = {}) {\n    super();\n    this.adapter = adapter;\n    this.config = this.mergeDefaultConfig(config);\n  }\n\n  async complete(prompt: string, options: AICompletionOptions = {}): Promise<AIResponse> {\n    this.emit('response:start', { prompt });\n\n    try {\n      const messages = this.prepareMessages(prompt);\n      const cacheKey = this.getCacheKey(messages);\n\n      // Check cache for consistent responses\n      if (this.config.consistencyLevel === 'high' && this.cache.has(cacheKey)) {\n        const cachedResponse = this.cache.get(cacheKey)!;\n        this.emit('response:complete', { response: cachedResponse });\n        return cachedResponse;\n      }\n\n      const completionOptions = this.enhanceOptions(options);\n      const response = await this.adapter.complete(messages, completionOptions);\n      \n      // Cache the response\n      if (this.config.consistencyLevel !== 'low') {\n        this.cache.set(cacheKey, response);\n      }\n\n      // Store in undo history\n      if (this.config.allowUndo) {\n        this.addToUndoHistory(prompt, response);\n      }\n\n      // Apply pacing if needed\n      await this.applyPacing();\n\n      this.emit('response:complete', { response });\n      return response;\n    } catch (error) {\n      this.emit('response:error', { error: error as Error });\n      throw error;\n    }\n  }\n\n  async *stream(prompt: string, options: AICompletionOptions = {}): AsyncIterable<AIStreamChunk> {\n    if (!this.adapter.stream) {\n      throw new Error(`${this.adapter.name} adapter does not support streaming`);\n    }\n\n    this.emit('response:start', { prompt });\n\n    try {\n      const messages = this.prepareMessages(prompt);\n      const completionOptions = this.enhanceOptions(options);\n      \n      let fullResponse = '';\n      const startTime = Date.now();\n\n      for await (const chunk of this.adapter.stream(messages, completionOptions)) {\n        fullResponse += chunk.delta;\n        this.emit('response:chunk', { chunk });\n        yield chunk;\n\n        if (chunk.done) {\n          const response: AIResponse = {\n            content: fullResponse,\n            usage: chunk.usage,\n            finishReason: 'stop',\n            metadata: {\n              streamDuration: Date.now() - startTime,\n            },\n          };\n\n          // Store in undo history\n          if (this.config.allowUndo) {\n            this.addToUndoHistory(prompt, response);\n          }\n\n          this.emit('response:complete', { response });\n          break;\n        }\n      }\n    } catch (error) {\n      this.emit('response:error', { error: error as Error });\n      throw error;\n    }\n  }\n\n  updateConfig(newConfig: Partial<PredictableAIConfig>): void {\n    const previousConfig = { ...this.config };\n    this.config = { ...this.config, ...newConfig };\n    \n    // Clear cache if consistency level changed\n    if (newConfig.consistencyLevel && newConfig.consistencyLevel !== previousConfig.consistencyLevel) {\n      this.cache.clear();\n    }\n\n    this.emit('config:changed', { config: newConfig });\n  }\n\n  updateFromCognitivePreferences(preferences: CognitivePreferences): void {\n    const configUpdate: Partial<PredictableAIConfig> = {};\n\n    // Map reading speed to pacing\n    if (preferences.readingSpeed !== undefined) {\n      const pacingMap = {\n        slow: 'slow' as const,\n        medium: 'normal' as const,\n        fast: 'quick' as const,\n      };\n      configUpdate.pacing = pacingMap[preferences.readingSpeed];\n    }\n\n    // Map explanation level\n    if (preferences.explanationLevel !== undefined) {\n      const explanationMap = {\n        simple: 'simple' as const,\n        moderate: 'moderate' as const,\n        detailed: 'detailed' as const,\n      };\n      configUpdate.explanationLevel = explanationMap[preferences.explanationLevel];\n    }\n\n    // Map processing pace to consistency\n    if (preferences.processingPace !== undefined) {\n      const consistencyMap = {\n        relaxed: 'high' as const,\n        standard: 'moderate' as const,\n        quick: 'low' as const,\n      };\n      configUpdate.consistencyLevel = consistencyMap[preferences.processingPace];\n    }\n\n    this.updateConfig(configUpdate);\n  }\n\n  undo(): UndoState | null {\n    if (!this.config.allowUndo || this.undoHistory.length === 0) {\n      return null;\n    }\n\n    if (this.currentStep > 0) {\n      this.currentStep--;\n      const state = this.undoHistory[this.currentStep];\n      this.emit('undo:performed', { step: this.currentStep, state });\n      return state;\n    }\n\n    return null;\n  }\n\n  redo(): UndoState | null {\n    if (!this.config.allowUndo || this.currentStep >= this.undoHistory.length - 1) {\n      return null;\n    }\n\n    this.currentStep++;\n    const state = this.undoHistory[this.currentStep];\n    this.emit('undo:performed', { step: this.currentStep, state });\n    return state;\n  }\n\n  canUndo(): boolean {\n    return this.config.allowUndo && this.currentStep > 0;\n  }\n\n  canRedo(): boolean {\n    return this.config.allowUndo && this.currentStep < this.undoHistory.length - 1;\n  }\n\n  getUndoHistory(): UndoState[] {\n    return [...this.undoHistory];\n  }\n\n  clearUndoHistory(): void {\n    this.undoHistory = [];\n    this.currentStep = 0;\n  }\n\n  clearCache(): void {\n    this.cache.clear();\n  }\n\n  getConfig(): PredictableAIConfig {\n    return { ...this.config };\n  }\n\n  private prepareMessages(prompt: string): AIMessage[] {\n    const systemPrompt = this.createSystemPrompt();\n    const messages: AIMessage[] = [];\n\n    if (systemPrompt) {\n      messages.push({\n        role: 'system',\n        content: systemPrompt,\n        timestamp: Date.now(),\n      });\n    }\n\n    messages.push({\n      role: 'user',\n      content: prompt,\n      timestamp: Date.now(),\n    });\n\n    return messages;\n  }\n\n  private createSystemPrompt(): string {\n    const prompts: string[] = [];\n\n    // Tone guidance\n    const toneInstructions = {\n      'calm-supportive': 'Respond in a calm, supportive manner. Use gentle language and provide reassurance.',\n      'encouraging': 'Be encouraging and positive. Focus on possibilities and strengths.',\n      'neutral': 'Maintain a neutral, professional tone. Be clear and objective.',\n      'clinical': 'Use precise, clinical language. Be factual and methodical.',\n      'friendly': 'Be warm and friendly. Use conversational language while remaining helpful.',\n    };\n    prompts.push(toneInstructions[this.config.tone]);\n\n    // Explanation level guidance\n    const explanationInstructions = {\n      simple: 'Use simple language and basic concepts. Avoid jargon and complex terms.',\n      moderate: 'Provide balanced explanations with some detail. Use accessible language.',\n      detailed: 'Give comprehensive explanations with examples and context.',\n      technical: 'Use precise technical language and detailed explanations.',\n    };\n    prompts.push(explanationInstructions[this.config.explanationLevel]);\n\n    // Analogy guidance\n    if (this.config.useAnalogies) {\n      prompts.push('When helpful, use analogies and real-world examples to clarify concepts.');\n    }\n\n    // Pacing guidance\n    const pacingInstructions = {\n      slow: 'Take time to explain each step. Break down complex ideas into smaller parts.',\n      normal: 'Provide clear explanations at a steady pace.',\n      quick: 'Be concise and direct. Focus on key points.',\n    };\n    prompts.push(pacingInstructions[this.config.pacing]);\n\n    return prompts.join(' ');\n  }\n\n  private enhanceOptions(options: AICompletionOptions): AICompletionOptions {\n    const enhanced = { ...options };\n\n    // Set consistent temperature based on consistency level\n    if (enhanced.temperature === undefined) {\n      const temperatureMap = {\n        low: 0.8,\n        moderate: 0.5,\n        high: 0.2,\n      };\n      enhanced.temperature = temperatureMap[this.config.consistencyLevel];\n    }\n\n    return enhanced;\n  }\n\n  private addToUndoHistory(prompt: string, response: AIResponse): void {\n    const state: UndoState = {\n      step: this.currentStep + 1,\n      prompt,\n      response,\n      timestamp: Date.now(),\n    };\n\n    // Remove any redo steps when adding new state\n    if (this.currentStep < this.undoHistory.length - 1) {\n      this.undoHistory = this.undoHistory.slice(0, this.currentStep + 1);\n    }\n\n    this.undoHistory.push(state);\n    this.currentStep = this.undoHistory.length - 1;\n\n    // Limit undo history size\n    if (this.undoHistory.length > this.config.maxUndoSteps) {\n      this.undoHistory.shift();\n      this.currentStep--;\n    }\n  }\n\n  private getCacheKey(messages: AIMessage[]): string {\n    const content = messages.map(m => `${m.role}:${m.content}`).join('|');\n    const configKey = `${this.config.tone}:${this.config.explanationLevel}:${this.config.pacing}`;\n    return `${this.adapter.name}:${this.adapter.getModel()}:${configKey}:${content}`;\n  }\n\n  private async applyPacing(): Promise<void> {\n    if (this.config.pacing === 'slow') {\n      await new Promise(resolve => setTimeout(resolve, 500));\n    } else if (this.config.pacing === 'normal') {\n      await new Promise(resolve => setTimeout(resolve, 200));\n    }\n    // No delay for quick pacing\n  }\n\n  private mergeDefaultConfig(config: Partial<PredictableAIConfig>): PredictableAIConfig {\n    return {\n      tone: 'neutral',\n      explanationLevel: 'moderate',\n      pacing: 'normal',\n      consistencyLevel: 'moderate',\n      useAnalogies: true,\n      allowUndo: true,\n      maxUndoSteps: 10,\n      ...config,\n    };\n  }\n}","/**\n * Advanced Adaptive Engine - Real-time ML-powered personalization for neurodivergent users\n * @fileoverview Implements advanced machine learning algorithms for personalized accessibility adaptations\n */\n\nimport { EventEmitter } from 'eventemitter3';\nimport type { Preferences } from '@neuroadapt/core';\n\n// Define local interfaces for missing types\nexport interface CognitiveLoadMetrics {\n  cognitiveLoad: number;\n  taskComplexity: number;\n  memoryLoad: number;\n  distractorImpact: number;\n}\n\nexport interface SensoryData {\n  audioSensitivity?: number;\n  vibrationTolerance?: number;\n  lightSensitivity?: number;\n  temperaturePreference?: number;\n}\n\nexport interface AdaptationPattern {\n  id: string;\n  userId: string;\n  patterns: {\n    visual: VisualPattern;\n    cognitive: CognitivePattern;\n    motor: MotorPattern;\n    sensory: SensoryPattern;\n  };\n  confidence: number;\n  lastUpdated: Date;\n  effectivenessScore: number;\n}\n\nexport interface VisualPattern {\n  preferredContrast: number;\n  colorSensitivity: string[];\n  motionTolerance: number;\n  fontSizePreference: number;\n  brightnessAdaptation: number;\n}\n\nexport interface CognitivePattern {\n  processingSpeed: number;\n  workingMemoryCapacity: number;\n  attentionSpan: number;\n  preferredInformationDensity: number;\n  distractionSensitivity: number;\n}\n\nexport interface MotorPattern {\n  clickAccuracy: number;\n  hoverDuration: number;\n  keyboardSpeed: number;\n  preferredTargetSize: number;\n  gestureComplexity: number;\n}\n\nexport interface SensoryPattern {\n  soundSensitivity: number;\n  vibrationTolerance: number;\n  lightSensitivity: number;\n  temperaturePreference: number;\n  texturePreference: string[];\n}\n\nexport interface AdaptationEvent {\n  type: 'visual_adjustment' | 'cognitive_load_change' | 'motor_assistance' | 'sensory_calibration';\n  data: any;\n  confidence: number;\n  timestamp: Date;\n}\n\nexport interface MLModelConfig {\n  modelType: 'neural_network' | 'decision_tree' | 'random_forest' | 'svm';\n  parameters: Record<string, any>;\n  trainingDataSize: number;\n  accuracy: number;\n  lastTrained: Date;\n}\n\n/**\n * Advanced Adaptive Engine using Machine Learning for real-time personalization\n */\nexport class AdaptiveEngine extends EventEmitter {\n  private patterns: Map<string, AdaptationPattern> = new Map();\n  private mlModels: Map<string, MLModelConfig> = new Map();\n  private realTimeMetrics: Map<string, any[]> = new Map();\n  private adaptationHistory: AdaptationEvent[] = [];\n  private isLearning: boolean = true;\n  \n  constructor(\n    private config: {\n      learningRate: number;\n      confidenceThreshold: number;\n      maxPatternAge: number;\n      realTimeAdaptation: boolean;\n    } = {\n      learningRate: 0.1,\n      confidenceThreshold: 0.8,\n      maxPatternAge: 30 * 24 * 60 * 60 * 1000, // 30 days\n      realTimeAdaptation: true\n    }\n  ) {\n    super();\n    this.initializeModels();\n  }\n\n  /**\n   * Initialize machine learning models for different adaptation types\n   */\n  private initializeModels(): void {\n    // Visual adaptation model\n    this.mlModels.set('visual', {\n      modelType: 'neural_network',\n      parameters: {\n        layers: [64, 32, 16],\n        activation: 'relu',\n        learningRate: 0.001,\n        epochs: 100\n      },\n      trainingDataSize: 0,\n      accuracy: 0,\n      lastTrained: new Date()\n    });\n\n    // Cognitive load prediction model\n    this.mlModels.set('cognitive', {\n      modelType: 'random_forest',\n      parameters: {\n        n_estimators: 100,\n        max_depth: 10,\n        min_samples_split: 5\n      },\n      trainingDataSize: 0,\n      accuracy: 0,\n      lastTrained: new Date()\n    });\n\n    // Motor assistance model\n    this.mlModels.set('motor', {\n      modelType: 'svm',\n      parameters: {\n        kernel: 'rbf',\n        C: 1.0,\n        gamma: 'scale'\n      },\n      trainingDataSize: 0,\n      accuracy: 0,\n      lastTrained: new Date()\n    });\n\n    // Sensory adaptation model\n    this.mlModels.set('sensory', {\n      modelType: 'decision_tree',\n      parameters: {\n        max_depth: 15,\n        min_samples_leaf: 3,\n        criterion: 'gini'\n      },\n      trainingDataSize: 0,\n      accuracy: 0,\n      lastTrained: new Date()\n    });\n  }\n\n  /**\n   * Analyze user behavior and generate adaptation patterns\n   */\n  async analyzeUserBehavior(\n    userId: string,\n    interactions: any[],\n    preferences: Preferences,\n    cognitiveMetrics: CognitiveLoadMetrics,\n    sensoryData: SensoryData\n  ): Promise<AdaptationPattern> {\n    const existingPattern = this.patterns.get(userId);\n    \n    // Analyze visual patterns\n    const visualPattern = await this.analyzeVisualPatterns(interactions, preferences);\n    \n    // Analyze cognitive patterns\n    const cognitivePattern = await this.analyzeCognitivePatterns(cognitiveMetrics, interactions);\n    \n    // Analyze motor patterns\n    const motorPattern = await this.analyzeMotorPatterns(interactions);\n    \n    // Analyze sensory patterns\n    const sensoryPattern = await this.analyzeSensoryPatterns(sensoryData, preferences);\n    \n    // Calculate confidence based on data quality and consistency\n    const confidence = this.calculatePatternConfidence(\n      visualPattern,\n      cognitivePattern,\n      motorPattern,\n      sensoryPattern,\n      interactions.length\n    );\n    \n    // Calculate effectiveness score\n    const effectivenessScore = await this.calculateEffectivenessScore(\n      userId,\n      { visual: visualPattern, cognitive: cognitivePattern, motor: motorPattern, sensory: sensoryPattern }\n    );\n    \n    const pattern: AdaptationPattern = {\n      id: `pattern_${userId}_${Date.now()}`,\n      userId,\n      patterns: {\n        visual: visualPattern,\n        cognitive: cognitivePattern,\n        motor: motorPattern,\n        sensory: sensoryPattern\n      },\n      confidence,\n      lastUpdated: new Date(),\n      effectivenessScore\n    };\n    \n    this.patterns.set(userId, pattern);\n    \n    // Trigger real-time adaptations if enabled\n    if (this.config.realTimeAdaptation && confidence > this.config.confidenceThreshold) {\n      await this.applyRealTimeAdaptations(pattern);\n    }\n    \n    this.emit('pattern_updated', pattern);\n    return pattern;\n  }\n\n  /**\n   * Analyze visual interaction patterns\n   */\n  private async analyzeVisualPatterns(interactions: any[], preferences: Preferences): Promise<VisualPattern> {\n    const visualInteractions = interactions.filter(i => i.type === 'visual');\n    \n    // ML-based analysis of visual preferences\n    const contrastAnalysis = this.analyzeContrastPreferences(visualInteractions);\n    const colorAnalysis = this.analyzeColorSensitivity(visualInteractions, preferences);\n    const motionAnalysis = this.analyzeMotionTolerance(visualInteractions);\n    const fontAnalysis = this.analyzeFontPreferences(visualInteractions);\n    const brightnessAnalysis = this.analyzeBrightnessAdaptation(visualInteractions);\n    \n    return {\n      preferredContrast: contrastAnalysis.optimal,\n      colorSensitivity: colorAnalysis.sensitivities,\n      motionTolerance: motionAnalysis.tolerance,\n      fontSizePreference: fontAnalysis.optimalSize,\n      brightnessAdaptation: brightnessAnalysis.adaptation\n    };\n  }\n\n  /**\n   * Analyze cognitive load patterns using ML\n   */\n  private async analyzeCognitivePatterns(metrics: CognitiveLoadMetrics, interactions: any[]): Promise<CognitivePattern> {\n    const cognitiveModel = this.mlModels.get('cognitive');\n    \n    // Process cognitive metrics through ML model\n    const processingSpeed = this.predictProcessingSpeed(metrics, interactions);\n    const workingMemory = this.assessWorkingMemoryCapacity(metrics, interactions);\n    const attentionSpan = this.calculateAttentionSpan(interactions);\n    const informationDensity = this.optimizeInformationDensity(metrics, interactions);\n    const distractionSensitivity = this.assessDistractionSensitivity(metrics, interactions);\n    \n    return {\n      processingSpeed,\n      workingMemoryCapacity: workingMemory,\n      attentionSpan,\n      preferredInformationDensity: informationDensity,\n      distractionSensitivity\n    };\n  }\n\n  /**\n   * Analyze motor interaction patterns\n   */\n  private async analyzeMotorPatterns(interactions: any[]): Promise<MotorPattern> {\n    const motorInteractions = interactions.filter(i => i.type === 'motor');\n    \n    return {\n      clickAccuracy: this.calculateClickAccuracy(motorInteractions),\n      hoverDuration: this.calculateOptimalHoverDuration(motorInteractions),\n      keyboardSpeed: this.calculateKeyboardSpeed(motorInteractions),\n      preferredTargetSize: this.calculateOptimalTargetSize(motorInteractions),\n      gestureComplexity: this.assessGestureComplexity(motorInteractions)\n    };\n  }\n\n  /**\n   * Analyze sensory adaptation patterns\n   */\n  private async analyzeSensoryPatterns(sensoryData: SensoryData, preferences: Preferences): Promise<SensoryPattern> {\n    return {\n      soundSensitivity: sensoryData.audioSensitivity || 0.5,\n      vibrationTolerance: sensoryData.vibrationTolerance || 0.5,\n      lightSensitivity: sensoryData.lightSensitivity || 0.5,\n      temperaturePreference: sensoryData.temperaturePreference || 22,\n      texturePreference: preferences.sensory?.texturePreferences || []\n    };\n  }\n\n  /**\n   * Apply real-time adaptations based on patterns\n   */\n  private async applyRealTimeAdaptations(pattern: AdaptationPattern): Promise<void> {\n    const adaptations: AdaptationEvent[] = [];\n    \n    // Visual adaptations\n    if (pattern.patterns.visual.motionTolerance < 0.3) {\n      adaptations.push({\n        type: 'visual_adjustment',\n        data: { reduceMotion: true, animationDuration: 0 },\n        confidence: pattern.confidence,\n        timestamp: new Date()\n      });\n    }\n    \n    // Cognitive load adaptations\n    if (pattern.patterns.cognitive.processingSpeed < 0.5) {\n      adaptations.push({\n        type: 'cognitive_load_change',\n        data: { \n          simplifyInterface: true, \n          reduceInformationDensity: 0.3,\n          increasePausesBetweenActions: 500 \n        },\n        confidence: pattern.confidence,\n        timestamp: new Date()\n      });\n    }\n    \n    // Motor assistance adaptations\n    if (pattern.patterns.motor.clickAccuracy < 0.7) {\n      adaptations.push({\n        type: 'motor_assistance',\n        data: { \n          increaseTargetSize: 1.5,\n          enableClickAssistance: true,\n          reduceFineMotorRequirements: true \n        },\n        confidence: pattern.confidence,\n        timestamp: new Date()\n      });\n    }\n    \n    // Sensory calibrations\n    if (pattern.patterns.sensory.soundSensitivity > 0.8) {\n      adaptations.push({\n        type: 'sensory_calibration',\n        data: { \n          muteNonEssentialSounds: true,\n          reduceAudioVolume: 0.3,\n          enableVisualAlternatives: true \n        },\n        confidence: pattern.confidence,\n        timestamp: new Date()\n      });\n    }\n    \n    // Apply adaptations\n    for (const adaptation of adaptations) {\n      this.adaptationHistory.push(adaptation);\n      this.emit('adaptation_applied', adaptation);\n    }\n  }\n\n  /**\n   * Train ML models with new data\n   */\n  async trainModels(trainingData: any[]): Promise<void> {\n    for (const [modelName, model] of this.mlModels.entries()) {\n      const modelData = trainingData.filter(d => d.type === modelName);\n      \n      if (modelData.length > 10) { // Minimum data requirement\n        // Simulate ML training (in real implementation, this would use actual ML libraries)\n        const accuracy = await this.simulateModelTraining(model, modelData);\n        \n        model.accuracy = accuracy;\n        model.trainingDataSize = modelData.length;\n        model.lastTrained = new Date();\n        \n        this.emit('model_trained', { modelName, accuracy, dataSize: modelData.length });\n      }\n    }\n  }\n\n  /**\n   * Predict optimal adaptations for new user interactions\n   */\n  async predictAdaptations(\n    userId: string,\n    currentContext: any\n  ): Promise<AdaptationEvent[]> {\n    const pattern = this.patterns.get(userId);\n    if (!pattern || pattern.confidence < this.config.confidenceThreshold) {\n      return [];\n    }\n    \n    const predictions: AdaptationEvent[] = [];\n    \n    // Use ML models to predict needed adaptations\n    for (const [modelType, model] of this.mlModels.entries()) {\n      if (model.accuracy > 0.7) {\n        const prediction = await this.runModelPrediction(model, currentContext, pattern);\n        if (prediction.confidence > this.config.confidenceThreshold) {\n          predictions.push(prediction);\n        }\n      }\n    }\n    \n    return predictions;\n  }\n\n  /**\n   * Get adaptation effectiveness metrics\n   */\n  getAdaptationMetrics(userId: string): {\n    totalAdaptations: number;\n    averageConfidence: number;\n    effectivenessScore: number;\n    lastAdaptation: Date | null;\n  } {\n    const userAdaptations = this.adaptationHistory.filter(a => \n      this.patterns.get(userId)?.id === a.data?.patternId\n    );\n    \n    const pattern = this.patterns.get(userId);\n    \n    return {\n      totalAdaptations: userAdaptations.length,\n      averageConfidence: userAdaptations.reduce((sum, a) => sum + a.confidence, 0) / userAdaptations.length || 0,\n      effectivenessScore: pattern?.effectivenessScore || 0,\n      lastAdaptation: userAdaptations.length > 0 ? userAdaptations[userAdaptations.length - 1].timestamp : null\n    };\n  }\n\n  // Private utility methods for ML analysis\n  private analyzeContrastPreferences(interactions: any[]): { optimal: number } {\n    // Simulate ML analysis\n    const contrastValues = interactions.map(i => i.contrastLevel || 1.0);\n    return { optimal: contrastValues.reduce((a, b) => a + b, 0) / contrastValues.length || 1.0 };\n  }\n\n  private analyzeColorSensitivity(interactions: any[], preferences: Preferences): { sensitivities: string[] } {\n    // Simulate color sensitivity analysis\n    return { sensitivities: preferences.visual?.colorBlindnessType ? [preferences.visual.colorBlindnessType] : [] };\n  }\n\n  private analyzeMotionTolerance(interactions: any[]): { tolerance: number } {\n    // Simulate motion tolerance analysis\n    const motionEvents = interactions.filter(i => i.type === 'motion_response');\n    return { tolerance: motionEvents.length > 0 ? motionEvents[0].tolerance || 0.5 : 0.5 };\n  }\n\n  private analyzeFontPreferences(interactions: any[]): { optimalSize: number } {\n    // Simulate font preference analysis\n    return { optimalSize: 16 }; // Default optimal size\n  }\n\n  private analyzeBrightnessAdaptation(interactions: any[]): { adaptation: number } {\n    // Simulate brightness adaptation analysis\n    return { adaptation: 0.8 }; // Default adaptation level\n  }\n\n  private predictProcessingSpeed(metrics: CognitiveLoadMetrics, interactions: any[]): number {\n    // Simulate processing speed prediction using ML\n    return metrics.taskComplexity ? 1 - metrics.taskComplexity : 0.7;\n  }\n\n  private assessWorkingMemoryCapacity(metrics: CognitiveLoadMetrics, interactions: any[]): number {\n    // Simulate working memory assessment\n    return metrics.memoryLoad ? 1 - metrics.memoryLoad : 0.7;\n  }\n\n  private calculateAttentionSpan(interactions: any[]): number {\n    // Calculate attention span from interaction patterns\n    return 30; // Default 30 seconds\n  }\n\n  private optimizeInformationDensity(metrics: CognitiveLoadMetrics, interactions: any[]): number {\n    // Optimize information density based on cognitive load\n    return metrics.cognitiveLoad ? Math.max(0.1, 1 - metrics.cognitiveLoad) : 0.7;\n  }\n\n  private assessDistractionSensitivity(metrics: CognitiveLoadMetrics, interactions: any[]): number {\n    // Assess sensitivity to distractions\n    return metrics.distractorImpact || 0.5;\n  }\n\n  private calculateClickAccuracy(interactions: any[]): number {\n    // Calculate click accuracy from motor interactions\n    const clicks = interactions.filter(i => i.action === 'click');\n    const successfulClicks = clicks.filter(c => c.successful);\n    return clicks.length > 0 ? successfulClicks.length / clicks.length : 1.0;\n  }\n\n  private calculateOptimalHoverDuration(interactions: any[]): number {\n    // Calculate optimal hover duration\n    const hovers = interactions.filter(i => i.action === 'hover');\n    const durations = hovers.map(h => h.duration || 500);\n    return durations.reduce((a, b) => a + b, 0) / durations.length || 500;\n  }\n\n  private calculateKeyboardSpeed(interactions: any[]): number {\n    // Calculate keyboard interaction speed\n    return 0.8; // Default speed factor\n  }\n\n  private calculateOptimalTargetSize(interactions: any[]): number {\n    // Calculate optimal target size based on accuracy\n    return 44; // Default 44px (WCAG recommendation)\n  }\n\n  private assessGestureComplexity(interactions: any[]): number {\n    // Assess comfortable gesture complexity level\n    return 0.5; // Default medium complexity\n  }\n\n  private calculatePatternConfidence(\n    visual: VisualPattern,\n    cognitive: CognitivePattern,\n    motor: MotorPattern,\n    sensory: SensoryPattern,\n    dataPoints: number\n  ): number {\n    // Calculate overall confidence based on data quality and consistency\n    const dataQualityScore = Math.min(1.0, dataPoints / 100); // More data = higher confidence\n    const patternConsistencyScore = 0.8; // Simulate consistency analysis\n    \n    return (dataQualityScore + patternConsistencyScore) / 2;\n  }\n\n  private async calculateEffectivenessScore(\n    userId: string,\n    patterns: { visual: VisualPattern; cognitive: CognitivePattern; motor: MotorPattern; sensory: SensoryPattern }\n  ): Promise<number> {\n    // Simulate effectiveness score calculation\n    return 0.85; // Default high effectiveness\n  }\n\n  private async simulateModelTraining(model: MLModelConfig, data: any[]): Promise<number> {\n    // Simulate ML model training and return accuracy\n    await new Promise(resolve => setTimeout(resolve, 100)); // Simulate training time\n    return Math.random() * 0.3 + 0.7; // Random accuracy between 0.7-1.0\n  }\n\n  private async runModelPrediction(\n    model: MLModelConfig,\n    context: any,\n    pattern: AdaptationPattern\n  ): Promise<AdaptationEvent> {\n    // Simulate ML model prediction\n    return {\n      type: 'visual_adjustment',\n      data: { prediction: 'optimize_contrast' },\n      confidence: model.accuracy,\n      timestamp: new Date()\n    };\n  }\n}\n\nexport default AdaptiveEngine; ","/**\n * Neural Adaptation System - Real-time neural network for accessibility adaptation\n * @fileoverview Implements neural networks for continuous learning and optimization of user preferences\n */\n\nimport { EventEmitter } from 'eventemitter3';\n\nexport interface NeuralLayer {\n  weights: number[][];\n  biases: number[];\n  activation: ActivationFunction;\n  dropout?: number;\n}\n\nexport type ActivationFunction = 'relu' | 'sigmoid' | 'tanh' | 'leaky_relu' | 'softmax';\n\nexport interface NeuralNetworkConfig {\n  layers: number[];\n  activations: ActivationFunction[];\n  learningRate: number;\n  momentum: number;\n  regularization: number;\n  batchSize: number;\n  maxEpochs: number;\n  convergenceThreshold: number;\n}\n\nexport interface TrainingData {\n  inputs: number[];\n  targets: number[];\n  userId: string;\n  timestamp: Date;\n  context: {\n    deviceType: string;\n    environment: string;\n    timeOfDay: number;\n    cognitiveState: number;\n  };\n}\n\nexport interface PredictionResult {\n  adaptations: {\n    visual: number[];\n    cognitive: number[];\n    motor: number[];\n    sensory: number[];\n  };\n  confidence: number;\n  reasoning: string[];\n  alternatives: Array<{\n    adaptations: any;\n    confidence: number;\n  }>;\n}\n\nexport interface LearningMetrics {\n  epoch: number;\n  loss: number;\n  accuracy: number;\n  validationLoss: number;\n  validationAccuracy: number;\n  learningRate: number;\n  convergenceRate: number;\n}\n\n/**\n * Neural Adaptation System for real-time accessibility optimization\n */\nexport class NeuralAdaptationSystem extends EventEmitter {\n  private network: NeuralLayer[] = [];\n  private trainingHistory: LearningMetrics[] = [];\n  private realtimeBuffer: TrainingData[] = [];\n  private isTraining: boolean = false;\n  private currentEpoch: number = 0;\n  \n  constructor(\n    private config: NeuralNetworkConfig = {\n      layers: [24, 64, 32, 16, 8], // Input -> Hidden layers -> Output\n      activations: ['relu', 'relu', 'relu', 'relu', 'sigmoid'],\n      learningRate: 0.001,\n      momentum: 0.9,\n      regularization: 0.01,\n      batchSize: 32,\n      maxEpochs: 1000,\n      convergenceThreshold: 0.001\n    }\n  ) {\n    super();\n    this.initializeNetwork();\n  }\n\n  /**\n   * Initialize neural network with random weights\n   */\n  private initializeNetwork(): void {\n    this.network = [];\n    \n    for (let i = 0; i < this.config.layers.length - 1; i++) {\n      const inputSize = this.config.layers[i];\n      const outputSize = this.config.l