Frankly.ai

Frankly.ai embeds a conversational AI agent directly within Microsoft Teams' native UI, leveraging Teams' conversation threading and message history APIs to maintain contextual awareness across multi-turn discussions. The system ingests Teams message objects (including metadata like sender, timestamp, thread depth) and uses this context to generate responses that reference prior messages and team dynamics without requiring users to manually copy-paste conversation history. Integration occurs via Teams Bot Framework and Graph API for message retrieval.

Frankly.ai implements data residency controls and compliance-aware filtering that prevents sensitive information (PII, regulated data) from being processed by external LLM providers or stored in non-compliant regions. The system uses pattern-matching and entity recognition to identify regulated data types (SSN, credit card, health records) and either redacts them before processing, routes requests to compliant regional endpoints, or blocks processing entirely based on organizational policy. This is implemented via pre-processing pipelines that run before LLM inference.

Frankly.ai implements scope-aware response generation where the AI understands which Teams channel, conversation, or team it's operating within and applies role-based access control (RBAC) to determine what information it can surface and what actions it can perform. The system uses Teams' native permission model (channel membership, team ownership, guest status) to enforce access boundaries, preventing the AI from surfacing confidential information to users without appropriate permissions. This is implemented via Teams Graph API permission checks before response generation.

Frankly.ai provides AI-assisted support workflow automation that analyzes incoming customer inquiries (via Teams messages or integrated ticketing systems) to automatically categorize tickets, suggest response templates, and identify escalation needs. The system uses text classification and intent recognition to route tickets to appropriate support tiers, generate draft responses based on historical resolution patterns, and flag urgent or complex issues for human review. This is implemented via NLP classification pipelines and retrieval-augmented generation (RAG) over historical support tickets.

Frankly.ai integrates with organizational knowledge bases (SharePoint, wikis, documentation) and uses retrieval-augmented generation (RAG) to ground AI responses in authoritative company information. The system embeds and indexes knowledge base documents, retrieves relevant passages based on customer inquiries, and generates responses that cite sources and maintain consistency with documented policies. This is implemented via vector embeddings (likely OpenAI or similar), semantic search over indexed documents, and prompt engineering to enforce citation and consistency.

Frankly.ai maintains conversation state across multiple turns within Teams threads, tracking context, user intent, and conversation history without requiring explicit state management by the developer. The system uses Teams' native message threading to persist conversation state, retrieves prior messages via Graph API on each turn, and maintains a working context window that includes relevant prior exchanges. This is implemented via Teams message history retrieval and in-memory context management with optional persistence to Azure storage.

Frankly.ai supports secure function calling and API integration with Microsoft ecosystem services (Dynamics 365, Power Automate, SharePoint, Azure services) via OAuth 2.0 and managed connectors. The system allows the AI to invoke business logic, retrieve data, or trigger workflows without exposing API keys or credentials, using Teams' identity context to authenticate API calls. This is implemented via Power Automate connectors, Azure Managed Identity, and secure credential storage in Azure Key Vault.

Frankly.ai provides comprehensive audit logging of all AI-assisted interactions, including what data was processed, what responses were generated, who reviewed/approved them, and what actions were taken. The system logs interactions to Azure storage with immutable audit trails, generates compliance reports for regulatory audits, and provides dashboards for monitoring AI usage patterns. This is implemented via structured logging to Azure Monitor/Application Insights and compliance report generation templates.

Frankly.ai analyzes customer sentiment in support conversations using NLP-based sentiment classification and detects escalation signals (frustration, urgency, threats) that indicate a ticket needs human intervention. The system scores sentiment on each message, tracks sentiment trends across the conversation, and flags conversations that show negative sentiment progression or explicit escalation indicators. This is implemented via pre-trained sentiment models and rule-based escalation heuristics.