cielo24 AI – human cognitive workflow engine – bridges the Artificial Intelligence Training and Accuracy gap by providing AI ground truth data and highly accurate media processing at scale.
The gap between proprietary data and workflow training makes it challenging for companies to adopt AI solutions.
AI accuracy is insufficient for many transcription, meta data and tagging use cases. As AI implementations grow, the need for human validation will increase. Humans are needed to seed AI algorithms, validate results, correct gaps, and measure accuracy.
As enterprises deploy AI, AI stacks must be customized for each corporate client and use case. The proprietary data and workflow training gap creates AI adoption friction.
cielo24 AI develops custom topic-based language models and training data, which help reduce barriers to AI adoption.
cielo24 AI media data platform processes millions of media minutes per month with a multi pass process using machines and 10Ks of humans. Clients and resellers include major online video platforms like Kaltura and Panopto, media companies like Reuters and WWE, and online publishers like Infobase and Skillsoft.
Market Opportunity and Trends
- Video needs data infrastructure for content and ad target
- 70%-80% accuracy for identifying people, brands, racism, profanity, etc., is insufficient
- Humans are necessary to seed AI algorithms, validate AI output, and measure accuracy
- Most companies utilize only machines (AI) or humans
- As AI grows, the need for human validation will grow…
Hybrid approach (machine first, human curation, and quality control passes)
Mix machines and humans at scale (10K to 100K minutes of media a day).
Workflows and tools identify audio and visual elements accurately.
AI ground truth to seed algorithms.
Project auditing to validate AI results.
Data warranty to determine metadata accuracy.
Dial up and down human involvement based on quality requirements.
As AI improves, more use cases and volumes.
Reduce human effort improving margins and scale.
Flexible configurations of machines and humans.
Multiple passes (machine 1st, human 2nd and 3rd).
Break video into chunks, process in parallel, and reassemble.
Approach increases speed and scale while reducing costs.
25K workers certified globally completing 50 to 75K tasks daily.
Pay by the quality task completed (Uber labor model).
Leverage freelance work platforms around the world.
Access to private workforces and 3rd party language partners.
Major platform companies in the entertainment, enterprise and education markets.
Automated end to end integrations.
Business partnerships include reseller and co-selling arrangements.