Conversational analytics and data exploration API.
DeepSearch is an AI-powered conversational analytics engine built by PayIntelli. It sits on top of a business's existing databases and allows any user — regardless of technical skill — to ask questions about their data in plain English and receive structured answers, charts, and narrative insights in real time. Users do not write SQL. They do not need to know how the database is structured. They ask a question, and DeepSearch does the rest.
Example: A payments operations manager asks, "Which payment providers had the highest decline rates last quarter, broken down by card type?" DeepSearch queries the database, runs the analysis, and returns a chart and a written summary — in seconds.
DeepSearch is delivered as a backend API service (WebSocket + REST), designed to be embedded into client-facing products. It is database-agnostic and supports multiple enterprise platforms out of the box.
The SQL bottleneck is universal. In most organisations, only a small number of engineers or analysts can query databases directly. Business users — operations teams, product managers, executives — must raise tickets, wait for reports, or rely on dashboards that rarely answer their specific question. This creates three compounding problems:
DeepSearch removes this dependency. Any authorised user can query their data conversationally, explore follow-up questions, and receive answers in the format most useful to them — without waiting for anyone.
At a high level, every DeepSearch interaction follows this flow:
The AI engine operates in an agentic loop — it can take multiple steps (discovering schema, running queries, generating visuals) before presenting a final answer. This mirrors how a skilled analyst would approach a complex question, but happens automatically and in seconds. All database credentials are stored encrypted. All queries are scoped to the authenticated user's data. The AI cannot access data belonging to other clients.
1. Conversational Analytics
The primary interface is a real-time WebSocket connection (the system maintains context across the conversation). Users can ask multi-turn questions — follow-ups, refinements, and comparisons — and the system maintains context across the conversation.
Key behaviours: - Ambiguity resolution: When a question is unclear (e.g., no date range specified), the AI asks a clarifying question and provides suggested answers the user can click. - Multi-query orchestration: Complex questions that require combining data from multiple sources are broken into parallel sub-queries, executed simultaneously, and synthesised into a single coherent answer. - Smart defaults: The AI uses today's date and known data patterns to infer reasonable defaults when the user has not specified them. - Honest boundaries: If the data needed to answer a question does not exist in the connected database, the AI says so rather than guessing.
2. Conversation History & Continuity
Every conversation is saved and retrievable: - Users can have up to 20 active conversations simultaneously - Conversations can be renamed, marked as favourite, and deleted - Full message history (questions, answers, charts) is available via REST API - Feedback (thumbs up / thumbs down) can be submitted per conversation - The AI uses recent conversation history as context, so follow-up questions ("now break that down by region") work naturally
Context is managed automatically — when a conversation grows very long, earlier turns are summarised to preserve the most relevant context without hitting AI model limits.
