SYSTEM ARCHITECTURE
Modules: Vision // Audio // Reasoning
Visual Layer
YOLOv5x6 + SmolVLM2
Custom object-detection pipeline tracks hands, vehicles, and text (OCR) across frames. Features synthetic thermal contrast via OpenCV INFERNO mapping for low-light situational awareness.
Audio Core
OpenAI Whisper Models
High-fidelity transcription aligned temporally with video footage. Allows officers to scrub video by navigating speech text, synchronizing radio and body-cam audio streams.
Reasoning Engine
Gemini 3 Pro
Constrained reasoning layer for querying incident records. Zero autonomous decision making—output is limited to summarization and navigation based strictly on present evidence.
Evidence Integrity & Security
Chain of Custody
Cryptographic hashing at upload creates a verifiable fingerprint. Raw footage is immutable—system outputs are stored as linked artifacts, never overwriting the source.
Privacy Glare
Redaction regions are tracked persistently. Data is restricted by default to agency-controlled environments. NO external model training on department data.