Learn more about Enabled Intelligence.
by Mark Tramontozzi
One of the things that sets Enabled Intelligence from our competitors is our hyperfocus on AI data quality – in particular the quality of the data used to train and optimize AI models and LLMs. Having worked on scores of mission critical AI projects for US government customers, we have seen repeatedly that seemingly small “upfront” differences in the quality of AI data inputs (i.e. how well AI training data is labeled) has huge, non-linear “downstream” impacts in AI model accuracy, performance, and compute cost. In other words, a 10% difference in AI training data quality can have a 50%, 60% or even higher impact on how well an AI model performs and on how much time and cost it takes to develop and run. And to help deliver superior AI data quality, Enabled Intelligence developed EnkI, our “quality-centric”, open-architecture ML/Ops software solution.
Why Did We Develop EnkI?
We did not set out to develop another ML/Ops platform – we wanted to leverage what already existed.
But in working with customers in the US defense and intelligence community who had extremely rigorous data quality requirements, we found most COTS labeling platforms and tools do not accurately or comprehensively capture real-time quality and performance metrics. So we developed EnkI for our customers – and they embraced it! EnkI manages the AI data labeling workflow from data ingestion to final label delivery. Included within EnkI is “EnkI Insights,” a fully integrated quality management and tracking tool that not only integrates data from multiple subsystems such as whatever data labeling software is being used but can also integrate data from a customer’s time accounting and human resource management systems. This latter capability is critically important because it allows customers to track both labeling productivity and labeling accuracy, and to identify and correct problems in real time.
EnkI automatically captures source metrics data and stores them in the EnkI Insights database. Customers can then view near real-time metrics using EnkI Insights’ customizable dashboards. With EnkI Insights, customers can drill down to view labeling performance at the team, individual annotator, quality analyst, and ontology level. No other ML/Ops platform that we are aware of provides this level of granularity.
With EnkI Insights, customers can measure metrics such as:
- Data assets completed per hour
- Data labels created per hour
- Overall progress to project completion
- Team and individual annotation error rates
- Predominant errors causing confusion for data annotators
- Missed label, i.e., non-identification of object, metadata classification, as required in ontology.
- False positive, i.e., misidentification of object or classification that is not present.
- Incorrect label, i.e., assigning incorrect metadata category, classification, or object class to annotation.
- Label precision and consistency
Concerned About AI Data Quality? Check Out EnkI
While EnkI was originally designed for mission-critical national security AI applications, EnkI can support the AI requirements for a wide range of applications, including LLM optimization and fine tuning, or the application of AI solutions for enterprises that have large amounts of data that need to be made “AI ready.” By using EnkI, customers can substantially “de-risk” their investments in AI technology by ensuring that the data being used in the AI model or application is high quality.
/
