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
hero-kurator
2.5M
Videos processed & indexed
~100×
Cost reduction vs. cloud-only
95%+
Transcription word accuracy
30K+
Hours of long-form content
the client
Kurator
Media & Entertainment Broadcast Footage Video Licensing
Kurator • Nimia
Video licensing & discovery

Kurator is a next-generation licensing and rights management platform designed for the media and entertainment industry. Originally founded in 2011 as Nimia, the company has over a decade of experience supporting leading organizations with licensing, rights clearance, and global content distribution.

Launched as Kurator in 2022, the platform modernizes this expertise into a cloud-based system that simplifies the buying, selling, and tracking of digital licenses for video and photo assets.

Trusted by industry leaders such as Netflix, HBO, NBC, ABC, AFP, TEGNA, and Terra Mater Studios, Kurator provides end-to-end solutions for license discovery, clearance, compliance, and payment tracking.

AccelOne partnered with Kurator to design and implement a robust, AI-powered video intelligence platform capable of transcribing, tagging, and enriching video content with time-encoded insights while drastically optimizing infrastructure costs.

the challenge

Business Challenge

Kurator faced several technical hurdles:

challenge 01

Scale

Managing millions of videos across diverse formats and qualities.

challenge 02

Metadata Gaps

Generating accurate, rich metadata (titles, summaries, descriptions).

challenge 03

Content Analysis

Identifying celebrities, brands, and contextual elements inside videos.

challenge 04

Searchability

Enabling time-encoded insights for precise navigation.

challenge 05

Hybrid Infrastructure

Balancing on-premises processing with cloud-based scalability.

challenge 06

Cost Efficiency

Cloud-based AI processing at scale was prohibitively expensive.

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.

infographic
Our Solution

Intelligent Video Processing Platform

AccelOne developed a platform with the following core features

core 01

Transcription 
Services

Accurate speech-to-text with time-stamped phrases.

Architecture
core 02

Scalable Video Management

Handling millions of videos in parallel.

Whisper (large)
core 03

Multi-Format Support

Seamless handling of resolutions from legacy to 8K.

1 frame / 2 sec
core 04

Content 
Analysis

Identifying people, brands, and contextual elements.

Gemma 3
core 05

Rich Metadata Generation

Titles, summaries, and descriptions automatically created.

OpenCV
core 06

Time-Coded 
Data

All insights are synchronized with video timelines.

50× fewer API calls
core 07

Hybrid AI Approach

Strategic mix of on-premises AI models and commercial APIs.

Architecture
the impact

Key Impact: Cost Reduction by Three Orders of Magnitude

One of the most significant achievements of this engagement was the reduction of AI processing costs by three orders of magnitude.

By combining on-premises transcription models with selective use of commercial APIs, AccelOne designed a hybrid architecture that:

This approach made the system not only scalable and secure but also economically viable at a massive scale, unlocking growth for Kurator without cost bottlenecks

01

Shifted the majority of workloads to locally run AI models (e.g., Whisper).

02

Leveraged APIs only where they added unique value (e.g., celebrity/brand recognition).

03

Minimized reliance on expensive cloud services for large-scale, repetitive tasks

the outcome

Results & Impact

~100×
Lower processing costs

Compared to cloud-only or traditional vendor pipelines.

~1000×
Lower processing costs

Massive reduction in processing costs for scale-intensive operations.

$2.5K–$3K
Per GPU machine

Scale throughput with low-cost infrastructure.

Scalable Video Intelligence

Transform video processing into a repeatable operational capability.

Infrastructure That Compounds

Add affordable GPU machines incrementally instead of committing.

Built for High Throughput

Designed to process large-scale video workloads efficiently.

Frequently asked questions

How did AccelOne help Kurator make millions of videos searchable and ready to purchase?
Kurator's core business is helping media buyers find and license the right moment inside long-form footage. With over 2.5 million videos and 30,000+ hours of content, that was impossible to do manually at scale. AccelOne built an AI pipeline that automatically transcribes, tags, and summarizes every video, indexing each one by what is said, what is shown, who appears, and when it happens. Buyers can now navigate directly to the relevant moment instead of scrubbing through hours of footage.
What made the AccelOne pipeline architecture different from standard cloud AI approaches?
The system was designed to avoid unnecessary work, not just process everything. Heavy video inference runs on on-premise GPU machines rather than cloud services, which eliminates the data transfer and compute costs that made cloud-only approaches unworkable at Kurator's scale. For tasks like celebrity detection, the pipeline runs a lightweight face check first and only calls AWS Rekognition when faces are actually present. Frames are also batched into mosaics before reaching external APIs, cutting those calls by up to 50x. The result is a system that scales without costs scaling with it.
What changed operationally for Kurator's team after the AI pipeline was deployed?
Before the pipeline, teams spent hours per batch manually entering transcripts, keywords, metadata, and compliance flags, often with inconsistent results. When large uploads came in, only partial tagging was applied, which hurt search quality and buyer confidence. After deployment, that process was reduced to a quick spot check and adding information that genuinely requires human judgment. The pipeline handles everything else automatically, consistently, and across the full catalog.
How accurate and reliable is the transcription powering Kurator's video search?
Transcription accuracy consistently exceeds 95% word accuracy in spot-checked samples, approaching human-level performance under good audio conditions. AccelOne configured Whisper (large) for English-only transcription to reach that threshold reliably. Those transcripts are the foundation of the entire search experience. They power keyword search across millions of assets, enable time-based navigation inside long videos, and feed the metadata extraction and summaries that make the catalog discoverable down to the exact moment.

Explore more perspectives from AccelOne