Case Study AI • Media & Entertainment

From Manual Metadata to Intelligent Video Search

Turning 2.5 million videos into searchable, time-addressable assets with a cost-efficient hybrid AI pipeline.

AI Pipeline Hybrid Architecture Video Intelligence
client: Kurator • Nimia Jan 2026
Group 5
2.5M
Videos processed & indexed
~100×
Cost reduction vs. cloud-only
95%+
Transcription word accuracy
30K+
Hours of long-form content
At a Glance
~0
Manual tagging remaining
Near-complete elimination.
hrs → min
Time saved per upload batch
From hours of manual tagging to minutes of quality review.
$2.5K
Cost per additional GPU unit
Scalable throughput by adding on-prem GPUs vs. six-figure contracts.
the client

A video discovery platform with millions of high-value assets

Kurator is a video licensing and discovery platform within Nimia, serving major media and entertainment buyers with high-value archival and broadcast footage — including news, interviews, and historical content.

The platform's core value is helping customers find the right moment inside long video assets, then enabling easy purchase with confidence in rights management. But at scale, that promise depended entirely on metadata quality.

the challenge

Search and tagging had become a serious bottleneck

With millions of videos, search and tagging had become a serious bottleneck. Several cloud-first and vendor-based approaches were explored but rejected due to cost, accuracy, and data transfer trade-offs that made them impractical at Kurator's scale.

challenge 01

Manual tagging didn't scale

Teams spent hours per batch entering transcripts, metadata, keywords, and compliance flags — often with inconsistent results across the catalog.

challenge 02

Manual tagging didn't scale

Teams spent hours per batch entering transcripts, metadata, keywords, and compliance flags — often with inconsistent results across the catalog.

challenge 01

Manual tagging didn't scale

Teams spent hours per batch entering transcripts, metadata, keywords, and compliance flags — often with inconsistent results across the catalog.

challenge 01

Manual tagging didn't scale

Teams spent hours per batch entering transcripts, metadata, keywords, and compliance flags — often with inconsistent results across the catalog.

the solution

A hybrid AI video intelligence pipeline built for scale

AccelOne designed and built a multi-model hybrid execution architecture that balances performance with economics — running heavy inference on-premises while using cloud services selectively and only when necessary.

archive 1
Results & impact

Real outcomes, measurable impact

Manual tagging is nearly eliminated, results are reliable enough for day-to-day production use, and the cost is dramatically lower than any traditional cloud-based approach.

~100×
Overall cost reduction

Compared to cloud-only or vendor pipelines. Throughput scales by adding low-cost GPU machines at $2.5K–$3K each — shifting video intelligence from a capital project into a repeatable operational capability.

95%+
Transcription word accuracy

Consistently exceeds 95% in spot-checked samples, approaching human-level performance under good audio conditions. Powers keyword search, time-based navigation, and downstream metadata extraction.

hrs→min
Reduction in tagging time

Manual metadata entry reduced from hours to minutes per batch. Teams perform a quick spot-check and add only information requiring human judgment, freeing them to focus on quality.

~1000×
Reduction in high-volume processing paths

In specific high-volume processing paths, including the 50× reduction in external API calls to AWS Rekognition via mosaic batching optimization.

Let’s Build What Works

Looking for a relevant example or similar engagement?
We’re happy to walk through comparable work in a short conversation. 

Let´s talk