Video Analysis Platform: Architecture and Capabilities
A video analysis platform goes beyond playback — it ingests video, applies AI inference, structures the results, and makes them queryable. VidForgeX is built for production workloads: multi-tenant, multi-vertical, with API-first design.
Platform architecture
VidForgeX is built on three infrastructure layers:
A managed database with Row-Level Security, a vector index for embeddings, and private, encrypted object storage for video and artifacts. All content partitioned by organization.
Serverless edge functions for API logic. Dedicated cloud VMs for AI-intensive analysis inference — routed per product line (Time & Motion, Forensic).
A global edge network for API proxying (api.vidforgex.com), MCP rate limiting (mcp.vidforgex.com) and frontend delivery.
Analysis routing and vertical separation
The analysis routing layer selects the correct inference VM for each job using a priority cascade:
- Enterprise org override — if the org has a dedicated endpoint configured, use it unconditionally
- Vertical pool — TMA, Forensic, and Sports each have their own dedicated VM pool
- Default shared pool — fallback for unrouted analysis types
This enables enterprise organizations to have completely dedicated compute infrastructure while the platform manages routing transparently.
Security and multi-tenancy
Every database query in VidForgeX executes through Row-Level Security policies that enforce org-scoped access. Each content table's policies require the row to belong to an organization the current user is a member of. There is no possibility of cross-org data access through the API layer.
Backend functions use a two-client pattern: an auth client for user authentication, and a service role client for data operations. This means all data access happens through elevated permissions only after user identity and org membership have been verified at the application layer.