Regulated AI & Compliance
NextPhase: Post-Market Surveillance for ML Models
Modality-agnostic, model-agnostic post-deployment ML monitoring
Post-market surveillance for machine learning: wrap any model, and NextPhase keeps watching its data and its performance once it is in production.
The problem
Models degrade quietly after deployment as data drifts, and most monitoring tools are tied to one framework or only handle tabular data.
The approach
A single wrapper logs every prediction against a training baseline, computes tabular drift (PSI, KS, Wasserstein) and text drift (embedding centroid, MMD, vocabulary), joins ground truth later for performance, and serves a FastAPI and React dashboard with rule-based alerts to Slack, email or webhooks.
The outcome
An open-source, model-agnostic monitoring toolkit in early alpha, with image support planned.
In detail
Designed and built NextPhase, an open-source Python library and dashboard for post-deployment (post-market) monitoring of ML models. A single wrapper turns any inference callable into a monitored model that logs inputs, outputs, latency and timestamps; a baseline from training data anchors drift detection, and ground-truth labels can be joined later for performance tracking. It covers tabular drift (PSI, KS, Wasserstein) and text drift (embedding centroid, MMD and vocabulary shift), with a FastAPI backend, a React dashboard of per-metric time series and an alert engine that notifies via webhook, Slack or email after consecutive breaching windows. Early alpha, under active development.
Built with
- MLOps
- Model Monitoring
- Drift Detection
- Python
- FastAPI
- React
More in Regulated AI & Compliance
- A 21 CFR Part 11 and ISO 13485 Compliant Electronic Document Management System
- An LLM-Assisted QMS That Auto-Scores Regulatory Readiness
Next project: A 21 CFR Part 11 and ISO 13485 Compliant Electronic Document Management System
By Kanishka Ranaweera, PhD, Machine Learning Engineer in Melbourne, Victoria, Australia. hello@kanishkabuilt.me