Corpora

Converts natural language questions into structured database queries through a conversational AI layer that interprets user intent and translates it to SQL or equivalent query syntax. The system maintains conversation context across multiple turns, allowing users to refine queries iteratively without re-specifying the full data context. This approach abstracts away query language complexity while preserving the ability to explore data through multi-turn dialogue.

Provides a visual interface to define custom conversational agents without requiring prompt engineering or code. Users configure bot behavior through form-based settings (system instructions, knowledge sources, response constraints) and the platform generates the underlying prompt templates and routing logic. This approach democratizes bot creation by abstracting prompt engineering complexity while maintaining customization through structured configuration rather than free-form text editing.

Automatically extracts patterns, trends, and actionable insights from conversation logs and query results through statistical analysis and LLM-based summarization. The system tracks which questions are asked most frequently, identifies data exploration patterns, and generates natural language summaries of key findings. This capability transforms raw interaction data into business intelligence without requiring manual analysis.

Connects to multiple data sources (databases, APIs, CSV uploads, cloud storage) and automatically infers or accepts schema definitions to enable unified querying across heterogeneous data. The system maintains a unified schema layer that maps source-specific field names and types to a canonical representation, allowing conversational queries to transparently span multiple sources. This abstraction enables users to query across silos without understanding underlying data structure differences.

Maintains conversation state and user context across multiple sessions, allowing bots to remember previous interactions, user preferences, and data exploration history. The system stores conversation metadata and relevant context in a session store (likely vector embeddings for semantic recall) and retrieves relevant prior context when answering new questions. This enables multi-session conversations where users can reference previous findings or continue exploratory analysis without re-establishing context.

Automatically formats query results and generates appropriate visualizations (charts, tables, summaries) based on result type and user context. The system infers visualization type from data shape (time series → line chart, categorical distribution → bar chart) and generates visualization specifications (Vega-Lite, Plotly, or similar) that can be rendered in the UI or exported. This capability makes data exploration more intuitive by presenting results in the most appropriate visual form without user configuration.

Allows users to upload or link documents, knowledge bases, or external sources that the bot uses as context for answering questions. The system ingests these sources, creates embeddings, and retrieves relevant passages during query execution to ground responses in provided knowledge. This enables bots to answer questions about specific datasets, documentation, or domain knowledge without requiring users to manually specify context in each query.

Caches frequently executed queries and their results to reduce latency and computational cost for repeated or similar queries. The system uses semantic similarity matching to identify when new queries are equivalent to cached results and returns cached data when appropriate. This optimization is transparent to users and improves performance for exploratory workflows where users often refine similar queries iteratively.

Implements configurable constraints on bot responses to prevent hallucinations, enforce data access policies, and ensure responses stay within defined boundaries. The system can restrict responses to knowledge sources only (preventing hallucinations), enforce data masking for sensitive fields, and validate responses against user-defined rules before returning them. This capability enables safe deployment of bots in regulated environments or with sensitive data.