Harmonic Security Optimizes Data Leakage Detection
Harmonic Security partnered with the AWS Generative AI Innovation Center to optimize its data leakage detection platform, fine-tuning a ModernBERT model using Amazon SageMaker AI, Amazon Bedrock, and Amazon Nova Pro. The team achieved sub-500 millisecond inference latency at the 95th percentile, maintained accuracy across PII, source code, and payroll categories, enabled EU-region deployment, and scaled with ml.g5.4xlarge endpoints and auto-scaling.
Key Points
- 1Reduced inference latency to sub-500 milliseconds at 95th percentile using fine-tuned ModernBERT and optimized inference
- 2Maintained detection accuracy for PII, source code, payroll while supporting EU data residency and auto-scaling
- 3Applied synthetic data from Llama 3.3 70B and Nova Pro, increasing training examples 10x–15x for balanced classes
Scoring Rationale
Practical, well-documented implementation delivering measurable latency and scalability gains; limited novelty beyond engineering optimizations for enterprise data-protection.
Sources
Public references used for this report.
Practice interview problems based on real data
1,625 SQL & Python problems across 15 industry datasets — the exact type of data you work with.
Try 250 free problems
