About

Bridging the data gap in chemistry & materials R&D

An integrated AI platform engineered by a dedicated team of researchers and developers to accelerate discovery across chemistry and materials R&D — from molecules to alloys, batteries, polymers, and formulations — validated against held-out data at every stage.

Our Mission

Traditional chemistry and materials discovery relies on slow, expensive trial-and-error experimentation. Matflow aims to change that by making small, sparse experimental datasets workable: every model is validated against held-out rows and public benchmark datasets before it is trusted, so the platform's recommendations carry honest, reproducible numbers rather than unverifiable promises.

4
Core Stages
5
Ensemble Models
20+
Evaluation Metrics
2
Public Benchmark Datasets

What Matflow is built for

Matflow is built for researchers who need defensible candidates, not black boxes: a no-code path from a CSV to a cost-ranked shortlist, with the math shown and every number reproducible from the benchmark page. It is developed by a dedicated team of researchers and developers, and validated against measured data — held-out rows, not assumptions.

References

The validation datasets and laboratory case study are documented in the platform's public benchmarks and evidence reports; the underlying methods are drawn from the peer-reviewed literature listed below. Links resolve through the DOI system.

High-throughput screening of cathode compositions

Published

Computational Materials Science, 197, 110592 (2021)doi ↗

Polymer structure–property relationships with machine learning

Published

Journal of Physics: Materials, 2(3), 034003 (2019)doi ↗

Phase stability in high-entropy alloys

Published

Materials Today Communications, 32, 104146 (2022)doi ↗

What we believe

Lab-grounded, not lab-adjacent

Every model and method is validated against measured data and public benchmark datasets before it is trusted — held-out rows, reproducible bundles, no black boxes.

Explainable by default

Researchers should never be told to trust a black box. Every prediction carries explanations and honest uncertainty.

Your data stays yours

Datasets, models and results are isolated per account and never pooled — no global models trained on your experiments.

Honest about limits

We state what the platform cannot do as plainly as what it can — extrapolation flags, small-data caveats and all.

The founder

Matflow began as CatForge — founded and built by one researcher with a simple conviction: small experimental data deserves big-science tooling.

Bassil Elshenawy

Founder

Bassil founded CatForge — now Matflow — as a founding contributor during his GenAI R&D internship at the Qatar Computing Research Institute (QCRI). He designed the platform's architectural core across synthesis, evaluation and prediction, and engineered most of what you use today: the full pipeline workflow, the integrated AI assistant, usage governance and the SHAP/LIME explainability views. He continues to advance the platform's science through generative-AI research at QCRI and QEERI — from continual model merging to LLM-driven 3D design — while studying AI engineering at Ulster University.

BEng Artificial Intelligence — Ulster UniversityGenAI Research — QCRIGenAI Research — QEERILLM & Generative AI systems

How far we've come

The platform ships in research cycles — here is the release history in brief.

Mar 2026
v1.0

Initial release: the four core stages — Data Synthesis, Evaluation, Prediction, Optimization.

Apr 2026
v1.5

TEA techno-economic analysis, Virtual Screening, and Design of Experiments (DoE).

May 2026
v2.0

Complete redesign, new UI layout, and the tiered access system.

Jun 2026
v2.2

Optimization constraints and filters, multi-objective search, and the active learning loop.

Jul 2026
v2.4

The Full Pipeline — stages chain end to end with TEA and synthetic-data nodes.

Aug 2026
v2.5

Platform redesign, blueprint architecture, and a new design system across every page.

The Matflow Unified Workflow

A rigorous four-stage workflow connecting raw experimental samples to synthesized candidates.

1

Data Synthesis

We deploy the Matflow synthesis engine — deep generative and copula-based models — to create high-fidelity synthetic data, expanding small experimental datasets while preserving complex non-linear correlations between variables.

2

Evaluation

Every synthetic dataset undergoes comprehensive validation using 20+ statistical metrics, including distribution match tests, correlation similarity scores, and distribution matching to ensure scientific reliability.

3

Prediction

Matflow's predictive ensemble — combining multiple tree-based and kernel-based regressors — is trained on the augmented data to act as a "what-if" simulator, with the Matflow Explainability Engine for transparent, interpretable predictions.

4

Optimization

The Matflow Optimizer Core — an adaptive, evolutionary search — navigates the vast parameter space of chemical compositions to discover novel candidates that maximize user-defined performance objectives. Generated candidates are ranked along a Pareto front, providing a portfolio of optimal designs that intelligently balance performance, stability, selectivity, and cost.

Powered by State-of-the-Art Technology

Cutting-edge machine learning and AI methodologies tailored specifically for physical chemistry.

Machine Learning

Matflow's multi-model ensemble methods for robust, accurate predictions.

Generative AI

The Matflow synthesis engine's neural, hybrid, deep and statistical modes for high-fidelity synthetic data generation.

Adaptive Optimization

The Matflow Optimizer Core for intelligent exploration and exploitation of chemical parameter spaces.

Explainable AI

The Matflow Explainability Engine (global and local impact) for transparent, interpretable model decisions and scientific insights.