AI Capabilities

Identifies and separates different speakers in recordings, labeling each segment with a speaker identifier. Approximately 90% accuracy. Critical for meeting transcription, interview analysis, and call processing attribution.

IDP combines multiple AI capabilities into an automated pipeline for documents and forms.

• OCR -- text extraction
• Layout detection -- document structure analysis
• Classification -- categorizing document type
• Data extraction -- pulling specific fields

Three OCR engines are available, configurable based on requirements and deployment model.

• PaddleOCR -- open-source, high accuracy
• Tesseract -- open-source, broad language support
• Azure Cognitive Services OCR -- cloud-based, best for complex layouts

Search across all content libraries in a single query. Instead of searching one folder at a time, the system searches all indexed videos, documents, images, and audio with permission-aware results ensuring data access control.

Queries across five data modalities simultaneously: transcripts, OCR text, detected objects, metadata, and visual descriptions. A single search returns results matching across any combination of these modalities ranked by relevance.

Content indexed in one language can be found through queries in another, leveraging 82-language transcription and translation. Transcription search includes time-stamped results for jumping directly to relevant moments.

Automated MD5 and SHA-256 checksum calculation at ingestion creates a cryptographic fingerprint. Any modification changes the hash, flagging alterations. This maintains chain of custody for legal and compliance workflows.

Exact duplicate detection via cryptographic hashing and near-duplicate detection via fuzzy logic algorithms. The workflow progresses from detection to manual review to policy-based action with full audit logging.

---

Connects to any IP camera or VMS supporting RTSP/ONVIF. The software receives video streams, runs AI detection in real time, and generates structured event records and alerts. No camera replacement or new hardware needed.

No need to replace existing cameras, VMS, or network infrastructure. AI Live Insight is a software layer adding AI analytics on top of existing equipment, eliminating capital expense and disruption of forklift upgrades.

Each AI Live Server instance (one GPU) processes 32 simultaneous camera streams. For larger deployments, add GPU server instances. The system scales horizontally from 5 cameras to hundreds.

Roll out incrementally starting with one corridor, district, or use case. Expand over time without deploying across all cameras at once. This reduces initial investment, proves value on a small scale, and enables budget-driven growth.

Several pretrained detection models work out of the box without training.

• Person detection
• Face detection and tracking
• Weapon detection (guns, knives, gunshots)
• License plate detection (ALPR/ANPR)
• PPE violation detection
• Vehicle detection (type, color)

AI Live Insight addresses 17 transit use cases validated with Toronto TTC.

• Fare evasion
• Suicide prevention
• Platform safeguarding
• Loitering
• Track intrusion
• Suspicious behavior
• Mental health incidents
• Fallen persons
• Mobility assistance
• Stranded customers
• Vandalism
• Trespassing
• Crowd counting
• Hazard detection
• Crowd movement
• Demographics
• Employee/customer distinction

PPE compliance detection, cycle-time analysis for production efficiency, and unsafe behavior detection on factory floors. Fine-tuned on customer-specific manufacturing environments with real-time alerts to safety managers.

AI Live Insight detects a range of smart city conditions, with all detections routing to 311/CRM systems.

• Copper wire theft activity
• Illegal dumping
• Road hazard and pothole detection
• Trash accumulation
• Double parking
• Bike-lane/ADA blockage
• Loading zone misuse
• Overstaying

AI Live Insight provides real-time public safety detection with alerting and historical playback.

• Crowd surge detection (stampede prevention)
• Abandoned vehicle detection
• Unattended object detection (IED prevention)
• Gunshot detection
• Activity detection (robbery, trespassing)

Every detection generates a structured record: timestamp, camera source, geographic location, detection type, classification, severity level, and confidence score. Records are stored, searchable, and exportable for investigation.

Alert routing is configurable per detection type -- weapon detection triggers immediate SMS while potholes generate 311 requests.

• Email
• SMS
• 311/CRM integration
• Traffic management systems
• Webhooks
• SCADA/operations platforms

Video capture triggers when AI detection occurs with configurable pre-event and post-event buffers. A weapon detection saves video from 30 seconds before through 60 seconds after, capturing context without continuous recording.