AI Capabilities

Intelligence Hub includes 103 features spanning the full spectrum of enterprise AI processing capabilities.

• Agentic RAG -- chatbot, agents, workflows, AI nodes
• AI Search -- 20+ search features
• Computer vision -- object detection, tracking, activity recognition
• NLP -- transcription, translation, sentiment, summarization
• Generative AI -- multi-LLM support, summarization, content generation
• Document intelligence -- OCR, layout detection, classification, data extraction
• File assessment -- integrity verification, deduplication
• Classification, anonymization, and data normalization

Intelligence Hub processes five categories of data via batch uploads, live feeds, and API pipelines.

• Video -- surveillance, body-worn camera, meetings
• Audio -- calls, recordings
• Images -- photos, scanned documents
• Documents -- PDFs, Word, spreadsheets, emails
• Other unstructured content -- any file type ingested through bulk or API

Intelligence Hub operates standalone, connecting with any data source via APIs. It can bundle with EnterpriseTube, VIDIZMO Digital Evidence Management System (DEMS), or Redactor for extended workflows, but this is optional, not required for core functionality.

A conversational AI system combining LLMs with retrieval that answers questions using your organization's actual data. Unlike standard chatbots, it retrieves content from documents, videos, and media before responding with source citations.

Agentic RAG goes beyond basic retrieval by adding autonomous reasoning and multi-step orchestration capabilities.

• Stateful multi-step workflows via LangGraph
• Multi-agent hierarchy with intent routing
• Conditional branching and iterative feedback loops
• Human-in-the-loop checkpoints
• Multi-modal retrieval across transcripts, objects, and metadata simultaneously

Retrieval spans five modalities in a single query, so a question can surface video clips, documents, and transcripts simultaneously.

• Transcripts -- 82-language speech-to-text
• OCR text -- extracted from documents and images
• Detected objects -- faces, vehicles, weapons
• Metadata -- tags, dates, locations
• AI-generated visual descriptions

A master bot analyzes user intent and routes to specialized child bots -- HR questions to an HR agent, IT questions to an IT agent. Each child has its own system prompt, scoped knowledge base, and configuration for domain-specific accuracy.

The chatbot supports four deployment channels, and agents can also embed in external systems through integration APIs.

• Web Widget -- embeddable in any website
• Google Chat -- as a bot
• Slack -- as a bot
• Portal -- within VIDIZMO's interface

Every response includes a confidence score. Configurable thresholds trigger automatic decline or human escalation when confidence drops below the set level. Human-in-the-loop checkpoints can be placed at any workflow step.

When the chatbot cannot answer queries satisfactorily, it flags these gaps for administrator notification. Admins receive alerts about topics needing more training data or content, enabling continuous knowledge base improvement.

A visual drag-and-drop graph editor for building multi-step AI workflows using LangGraph without writing code. Six node categories, flow control (IF/Loop/Switch/Parallel), triggers, and human review steps are all configurable.

Six node categories cover the full range of AI workflow automation needs.

• AI Nodes -- LLM with multi-provider support
• Embedding Nodes -- 8 providers
• Data Nodes -- JSON, PDF, web extractors
• Processing Nodes -- Python code, data operations
• Integration Nodes -- HTTP, MCP
• Flow Control -- IF, Loop, Switch, Parallel

Intelligence Hub supports multiple LLM providers concurrently, and Azure AI Foundry provides access to 1,200+ models for selection and fine-tuning.

• Azure OpenAI (GPT-4o)
• Google Gemini
• Anthropic Claude
• Ollama (self-hosted)
• VLLM (self-hosted)

Eight embedding providers are available, with local options supporting on-premises and air-gapped deployments with no external API calls.

• Anthropic
• Google
• HuggingFace (local)
• Infinity
• Ollama (local)
• OpenAI/Azure OpenAI
• VLLM API
• VLLM self-hosted

Self-hosted LLMs through Ollama and VLLM run entirely within customer infrastructure with no outbound connections. The same workflow designer and agent tools work with self-hosted models for classified or data-sovereign environments.

Each AI node can use a different model -- GPT-4o for summarization, Claude for analysis, and a self-hosted Ollama model for classification within the same workflow. This multi-model approach optimizes each task.

Combines vector search (semantic similarity), keyword search (text matching), and hierarchical indexing (structured navigation) for high recall and high precision. The ElasticSearch-based engine ensures comprehensive content discovery.

Intelligence Hub detects a wide range of objects across video, images, and live feeds with support for custom objects.

• People -- faces, persons, heads
• Vehicles -- cars, types, license plates
• Weapons -- guns, knives
• Electronics -- laptops, cellphones
• Documents/signs -- signatures, addresses
• Safety/PPE violations
• Environment -- graffiti
• Custom objects -- trainable per customer

Once detected, objects are tracked throughout the video with timestamp markers and frame-by-frame location data. This enables queries like "show everywhere this person appears" or timeline reconstruction for investigations.

Transformer-based temporal models (VideoMAE architecture) identify actions in video sequences with 92%+ accuracy. Shopping, trespassing, and robbery behaviors are detectable, plus organization-specific activity patterns configurable per customer.