Inside the V2 Platform Redesign

Sphot V2 is not a feature release. It is a full rebuild — every surface redesigned from first principles, every model pipeline replaced with a 19-stage production system, and a new internal operating layer (HQ) that manages the platform in real time.

This post covers the design decisions behind V2 and why they matter.

The Problem with V1

V1 was a working product, but it was built by layering features onto an architecture that was never designed for the volume or complexity we knew was coming. The pipeline was a single orchestrator with five stages. The analytics surface was a read-only dashboard. There was no studio mode, no client review portal, no moderation system.

The biggest problem: every piece was too tightly coupled. Changing the transcription model required touching six files. Adding a new platform required updating three schemas.

What Changed in V2

The Pipeline

V1 had 5 stages. V2 has 19 — each stage is an independent agent with a paired Sentinel critic that scores output 0–1 before the pipeline proceeds. No stage advances on a failing score. The pipeline is a film crew, not a relay race.

Platform Intelligence

Platform Intelligence is a new capability that analyses creator performance at the account level, identifies the patterns behind top-performing content, and encodes them into a creator's DNA profile. The profile feeds back into every generation decision — hook style, pacing, format.

The Creator Studio

The studio redesign was driven by a single constraint: the UI must not get in the way of the work. Every action has a keyboard shortcut. The timeline editor is non-destructive. Approvals have a dedicated queue mode that eliminates context switching.

HQ

HQ is the internal operating layer built for the Sphot team. It surfaces everything: pipeline health histograms, moderation queues, support tickets, audit logs, user management. Every operator action is logged. Every pipeline failure is surfaced. The cost of operating at scale is proportional to the difficulty of the job, not the volume of reels — that is the CVT.

What's Next

The nine wives. Each capability cluster — language, voice, vision, editing judgment, musicality, visual aesthetics, performance prediction, distribution intelligence, and the director — will be trained on production data. The product and the training program are the same machine.