Custom LLM for quality control
4 months
3 FTE: CTO, ML Engineer, Data Scientist
Project
K2 implemented custom LLM fine-tuned to correct mistakes and produce high quality metadata.
Results
Model developed by K2 helped the client achieve better data at no additional human cost. The solution's scalability allows ongoing development according to the client's evolving needs
MediaPress
In the dynamic realm of Media Press Group, our recent project focused on optimizing metadata creation and enrichment processes for television and streaming content. We fine-tuned a foundational LLM to empower the 500+ member editorial team, resulting in a 100% efficiency increase.
While maintaining a human supervision layer in the process, our monitoring dashboard demonstrated that model fine-tuning significantly reduced the need for human editors to correct created metadata, such as shortened synopses. The user-friendly interface ensures seamless collaboration, allowing non-technical users to actively refine the model tuning and further improve performance as needed.
This targeted approach, integrating skills such as data analytics and continuous improvement, not only addresses immediate needs but also positions Media Press Group as a leader in metadata management and distribution, showcasing a tangible boost to efficiency in metadata processes.
Delivering sustainable value.
Deliverables
Impact
Skills
Custom LLM fine-tuned on client data
Increased quality
Reduced risk of human mistakes
The 100% performance increase is a tangible indicator of our project's success.
This translates into not just numerical growth but a qualitative enhancement of the editorial processes, positioning the team for continued success.
Large Language Models
Fine-tuning