PIPELINE DEEP DIVE

Eight discrete stages with visible state and real dependency boundaries.

The Vetrra pipeline is designed for operators who want end-to-end visibility across ingestion, encode policy, subtitle handling, OCR, and final deployment. Each stage has explicit responsibilities and explicit failure surfaces.

STAGE MAP

Real tools. Deterministic flow. No generic automation language.

01

Intake

Receive source jobs from SABnzbd or manual operator queues and normalize metadata into one visible task list.

Primary concern: queue integrity and source classification.

02

Organize

Resolve naming, folder targets, path templates, and title normalization before heavy processing begins.

Primary concern: deterministic filesystem structure.

03

MKV processing

Inspect streams, validate track layout, and establish the container state that downstream stages inherit.

Primary concern: accurate container introspection.

04

Video encode

Apply hardware-aware FFmpeg profiles with operator-visible bitrate, ETA, acceleration, and fallback signaling.

Primary concern: quality and throughput policy.

05

Subtitle handling

Extract, validate, and classify subtitle streams, including language flags and forced-track rules.

Primary concern: subtitle policy and playback fidelity.

06

Muxing

Reassemble validated media tracks into the final container with explicit operator review surfaces.

Primary concern: container correctness.

07

Poster OCR

Use local Ollama vision models to parse poster text and keep media artwork classified with the rest of the library.

Primary concern: metadata extraction from visual assets.

08

QC and deploy

Validate final output structure, generate quality reports, and deploy the asset into its target library path.

Primary concern: trustworthy delivery.

DEPENDENCY MODEL

The application is local-first because the pipeline depends on local tooling.

FFmpeg

Handles encode, stream inspection, and hardware acceleration decisions surfaced directly in the operator UI.

MKVToolNix

Supports stream topology awareness, mux policy, and container integrity during processing and deployment.

SABnzbd

Feeds structured intake into the queue so acquisition and processing remain connected rather than fragmented.

Ollama vision models

Enable local OCR and poster parsing without shipping library data or artwork to cloud APIs.

POSTER HANDLING

Cinematic media belongs inside the operator shell, not in generic SaaS ornament.

The product proof section now frames media inside a faux workspace chrome that matches the application's shell geometry. Poster OCR, gallery handling, and diagnostics sit inside the same system language as the rest of the pipeline.

OLLAMA OCR BOUNDING BOXES CLASSIFIED
PIPELINE GALLERY Poster Proof
Poster wall preview used to demonstrate the OCR and gallery stage.

OPERATOR OUTCOME

Designed for home-lab operators, self-hosters, and data hoarders who want proof.

Vetrra does not hide complexity. It concentrates it into one interface so the operator can see where a job lives, which external dependency is active, what acceleration path is engaged, and why a final file is or is not ready to deploy.