MonEcho
HealthTech | SaaS
AI for prenatal ultrasound classification. where precision and regulatory compliance are non-negotiable.
MonEcho builds software for prenatal care practitioners. They wanted to explore whether AI could assist in ultrasound image classification. one of the most sensitive domains in healthcare, where accuracy directly impacts patient outcomes and regulatory requirements are strict.
Rakam built a complete exploration pipeline. fine-tuning with hyperparameter sweeps, zero-shot évaluation of foundational models, custom data stratification, and benchmarking. all delivered as Docker images with regulatory compliance built in from the start.
« MonEcho explores AI for one of the most sensitive domains in healthcare. prenatal ultrasound classification, where precision and compliance are non-negotiable. »
MonEcho
Client partnership
Fine-tune
Pipeline with sweeps
Zero-shot
Foundation models tested
Docker
Delivered as images
Compliant
Regulatory from day one
Business
Business Impact
MonEcho now has a validated AI exploration pipeline that demonstrates whether ultrasound classification is feasible with their data. This de-risks the decision to invest further in medical AI features, with clear benchmarks comparing fine-tuned and foundational model approaches.
Regulatory compliance was built in from day one. not retrofitted. The Docker-based delivery model ensures reproducibility and auditability, critical requirements for any medical device AI under European regulations.
Product
What We Built
Ultrasound Classifier
Image classification model for prenatal ultrasound images. Trained on domain-specific medical data with careful attention to class balance, edge cases, and clinical relevance of predictions.
Fine-Tuning Pipeline with Sweeps
Automated hyperparameter optimization pipeline that systematically explores model configurations. Sweeps test learning rates, architectures, augmentation strategies, and regularization to find optimal settings for medical imaging accuracy.
Zero-Shot Foundational Model évaluation
Benchmarking of state-of-the-art foundational vision models in zero-shot mode against fine-tuned approaches. Provides a clear comparison of accuracy, latency, and cost tradeoffs for MonEcho's specific use case.
Custom Data Stratification & Benchmarking
Domain-aware data stratification that ensures training, validation, and test splits respect clinical categories. Comprehensive benchmarking framework with metrics relevant to medical imaging. not just accuracy, but sensitivity and specificity per class.
Technical
Technical Architecture
The pipeline uses custom data stratification designed for medical imaging. ensuring clinically meaningful splits that prevent data leakage between patient studies. All experiments are tracked in MLflow for reproducibility.
Models are delivered as Docker images via OVH registry, ensuring consistent environments from training to inference. Kubernetes handles orchestration, and the entire pipeline is auditable for regulatory review.
# Stack
Custom data stratification
Docker / OVH Container Registry
MLflow (experiment tracking)
Kubernetes (orchestration)
# Approach
Fine-tuning with HP sweeps
Zero-shot foundational models
Medical-grade benchmarking
# Compliance
Regulatory-compliant from day 1
Reproducible Docker delivery
Auditable experiment tracking
Move from AI hype to the real thing
Leave us a message, our team will get back to you within 24 hours.