CovenAce — AI-Powered Covenant Management

Agentic AILegalTechOptimization

About the Product

CovenAce is Anaptyss's publicly featured AI-powered covenant management platform. It automates the extraction and validation of complex financial covenants from legal and credit agreements, significantly streamlining the compliance monitoring process for financial institutions.

My Role & Contributions

As a Lead AI Engineer, I was responsible for architecting the AI intelligence behind CovenAce. I worked directly with business stakeholders to understand the nuance of legal extraction and translated these requirements into a self-correcting multi-agent pipeline capable of high-accuracy document processing.

Technical Challenges Solved

  • Achieved 94.47% accuracy and 0.93 recall in entity classification across over 1,000 credit agreements, dealing with highly unstructured and variable text formats.
  • Optimized the inference pipeline using batch processing and parallel agent architectures, reducing extraction latency by 97% (from 5 hours to 10 minutes).
  • Designed a continuous prompt optimization framework utilizing aspect-based feedback, which improved F1-scores by 12% across diverse covenant types.

Technologies Used

Google ADKLLMsParallel ArchitecturePrompt OptimizationBatch Inference

Lessons Learned

The project demonstrated that in complex domains like LegalTech, static prompting is inadequate. Integrating feedback loops and self-reflection into the agentic workflows was key to consistently handling edge cases in legal phrasing.