Chat Whisperer

Automatically detects incoming user messages across 50+ languages and routes them to language-specific NLP pipelines, enabling seamless multilingual conversations without manual language selection. The system maintains separate conversation contexts per language thread, allowing users to switch languages mid-conversation while preserving conversation history and context. Implementation uses language identification models (likely fastText or similar) at message ingestion to classify input, then applies language-specific tokenization and response generation.

Provides a browser-based visual interface for designing chatbot conversation flows using node-and-edge graph abstractions, where non-technical users drag conversation nodes (user intents, bot responses, conditional branches) onto a canvas and connect them with decision logic. The builder compiles visual flows into an internal state machine representation that executes at runtime, supporting branching logic, variable interpolation, and integration points without requiring code. Architecture likely uses a graph-based workflow engine (similar to n8n or Zapier's visual builders) with JSON serialization of flow definitions.

Allows chatbot responses to include dynamic variables (e.g., {{customer_name}}, {{issue_type}}) that are replaced with actual values extracted from conversation context or user profile data at response generation time. The system extracts entities from user messages or retrieves user profile data, then substitutes variables in response templates with these values, enabling personalized responses without manual customization per user. Implementation uses a template engine (likely Handlebars, Jinja, or similar) that processes response templates with variable substitution.

Allows administrators to define chatbot tone, vocabulary, and response patterns through a configuration interface where they specify brand voice guidelines, response templates with variable interpolation, and personality traits that influence generated responses. The system applies these customizations at response generation time by injecting personality context into the LLM prompt or by selecting from curated response templates that match the defined brand voice. Implementation likely uses prompt engineering with personality descriptors or a template-matching system that ranks responses by tone alignment.

Maintains conversation state across multiple user messages within a session, storing message history, extracted entities (customer name, issue type), and conversation metadata in a session store. The system retrieves relevant context from previous messages when generating responses, enabling the chatbot to reference earlier statements and maintain coherent multi-turn conversations. Architecture uses session IDs to track conversations, likely with TTL-based expiration (e.g., 30-day session timeout) and optional persistence to a database for historical analysis.

Provides a web dashboard displaying aggregated metrics about chatbot conversations including message volume, conversation completion rates, average conversation length, and common user intents or topics. The system collects conversation metadata (duration, user satisfaction ratings if available, intent classification) and visualizes trends over time using basic charts and tables. Implementation likely uses event logging at message ingestion, aggregation in a time-series database, and visualization with a charting library (Chart.js, D3, or similar).

Enables Chat Whisperer chatbots to receive and send messages through external messaging platforms (likely Facebook Messenger, WhatsApp, Slack, or similar) by exposing webhook endpoints that accept incoming messages and providing API methods to send responses back to the originating platform. The system translates between Chat Whisperer's internal message format and each platform's API schema, handling platform-specific features like buttons, quick replies, or media attachments. Architecture uses a webhook listener that validates incoming requests, routes them to the chatbot engine, and calls the platform's send API with formatted responses.

Provides role-based access control (RBAC) for the Chat Whisperer admin dashboard, allowing account owners to create user accounts with different permission levels (admin, editor, viewer) that restrict access to chatbot configuration, analytics, and conversation data. The system authenticates users via email/password or SSO (if available) and enforces permissions at the API level, preventing unauthorized access to sensitive configuration or data. Implementation likely uses JWT tokens for session management and permission checks on each API endpoint.

Offers a free tier with limited chatbot instances, message volume, and feature access, designed to allow users to test Chat Whisperer's core functionality before committing to paid plans. The free tier likely includes 1-2 chatbot instances, a message quota (e.g., 1,000 messages/month), and access to basic features (no-code builder, multilingual support) while restricting advanced features (analytics, integrations, custom domains) to paid tiers. Implementation uses account-level feature flags and quota enforcement at message ingestion to limit free tier usage.

Implements encryption for conversation data at rest and in transit, protecting customer messages and personal information from unauthorized access. The system likely uses TLS/SSL for data in transit and AES-256 or similar for data at rest, with key management handled by Chat Whisperer's infrastructure. Implementation may include data residency options (e.g., EU data centers for GDPR compliance) and optional data retention policies that automatically delete old conversations.

Analyzes incoming user messages to classify them into predefined intent categories (e.g., 'billing question', 'technical support', 'product inquiry') and routes them to appropriate response handlers or escalation paths. The system uses NLP-based intent classification (likely a trained classifier or LLM-based classification) to match user input against known intents, then selects responses or actions based on the classified intent. Implementation may use a combination of keyword matching, semantic similarity, or a fine-tuned classifier, with fallback to a default 'unknown intent' handler.