Roadmap

Where Matflow is headed

The platform ships in research cycles, not marketing cycles. This page is the honest version of that: what you can use today, what is being built next, and what is on the horizon.

In the platform today

Pipeline

End-to-end chaining of Data Synthesis → Evaluation → Prediction → Optimization with synthetic-data nodes and TEA in the loop.

Cost-aware optimization

TEA with five costing modules and six-tenths scale-up rules feeding a single net-value objective.

Explainable prediction

Five-model ensemble with global/local impact, conformal intervals and extrapolation flags on every prediction.

Screening, DoE & active learning

Structural-similarity virtual screening, five design families, and a Bayesian active-learning loop.

AI Copilot + Data Enrichment

Page-aware AI Copilot grounded in the docs with inline citations, plus tabular extraction from literature PDFs.

Evaluation suite

Quality, ML-efficacy, dimensionality-reduction, Privacy Shield and anomaly-detection reports on every generated batch.

In the platform today — see the Changelog for release details.

Building next

Model registry & governance

Every trained model versioned, comparable, and promotable — no more retraining on every optimization run. Dataset versioning and lineage from raw upload to final candidate.

Model Hub expansion

New generative modes for Data Synthesis; new ensemble variants, Gaussian-process and stacking models for prediction; new multi-objective and Bayesian optimization strategies for search.

Formal privacy suite

DCR/NNDR metrics, membership-inference scores, and TSTR/TRTS efficacy — the standard set reviewers ask for — plus optional differential privacy for shareable synthetic data.

Deeper explainability

PDP/ICE plots, permutation importance, H-statistics, and quantile regression for asymmetric uncertainty.

Run templates & scheduling

Save any configuration as a reusable template, sweep parameters across batch runs, and re-optimize on a schedule as new lab data arrives.

Reproducibility bundles

Every run exports data, config, seed, environment and results as a single bundle a reviewer can reproduce.

In active planning — next major milestones.

On the horizon

Unit-aware columns

Declare units per column; the platform converts, validates dimensional consistency, and labels every axis correctly.

Chemical intelligence

Parse chemical formulas, look up elemental properties, and warn on impossible formulations before you generate.

Notifications

In-app, email and webhook events for run completion, quota levels, project invites and security events — with quiet hours.

Multilingual platform

English shipped with an Arabic locale scaffolded and full RTL support.

Public status history

90-day uptime bars, incident timelines and scheduled-maintenance notices on the status page — built on real recorded history, not estimates.

Organizations

Shared projects, role-based collaboration and comments — bringing multi-researcher groups into one workspace.

On the horizon — priorities may shift with research feedback.

Help shape what comes next

Feedback from researchers working on real data decides the roadmap — not speculation. Tell us what the platform is missing.