A web interface for AMLRO (Active Machine Learning Reaction Optimizer) that lets a bench chemist run a full active-learning reaction-optimization campaign: define a reaction scope, collect initial training data, then iteratively review AI-suggested reaction conditions, record results, and let the model retrain each cycle. No Python required.
Requires Python 3.10+ and Node.js 18+. Two terminals:
# Terminal 1: backend
cd backend
python -m venv .venv
.venv\Scripts\python -m pip install -e ".[dev]"
.venv\Scripts\python -m flask --app amlro_gui.app run --debug
# Terminal 2: frontend
cd frontend
npm install
npm run dev
Open http://127.0.0.1:5173. Commands above are for Windows; for macOS/Linux,
installing prerequisites, or what each command does, see
docs/GETTING_STARTED.md.
A reaction-optimization campaign has three phases, each with its own screen:
- Reaction scope: define continuous features (bounds, resolution) and categorical features (allowed values) and objectives (with min/max direction), or upload an existing dataset instead of starting fresh.
- Training: run the initial sampling plan AMLRO generates, recording an objective value for each condition.
- Prediction: each cycle, AMLRO trains a regression model on the data collected so far and suggests the next batch of conditions to try. Record results and repeat until the campaign is stopped.
Every experiment's state is file-backed, not stored in a session cookie, so closing the browser or restarting the app never loses progress, and past experiments can be resumed, listed, or removed from a landing page.
- Backend (
backend/): Flask, as a pure JSON API with no server-rendered templates, behind a thin routing layer, a Pydantic-validated schema layer, and a service layer that calls into AMLRO. State for each experiment is astate.jsonfile in that experiment's own directory. - Frontend (
frontend/): React and TypeScript, built with Vite, using Mantine for UI components, TanStack Query for API state, and Nivo for charts. - Served as one origin in production, with the built frontend as static files behind the Flask app.
docs/GETTING_STARTED.md: full setup/run instructions (Windows, macOS, Linux) and dev commands (lint, format, test).docs/ARCHITECTURE.md: code layout, request flow, and how to add a new resource following the existing pattern.
- AMLRO: the active-learning reaction-optimization engine this GUI wraps. An open-source framework that accelerates chemical reaction optimization using active learning with classical machine learning regression models, combining space-filling sampling strategies (Sobol, Latin Hypercube) with iterative model training, prediction, and experiment selection. It supports multiple regression models, multi-objective definitions, and user-defined parameter bounds for data-efficient optimization from small initial datasets. This repo has no optimization logic of its own; every reaction-scope, training, and prediction computation is delegated to AMLRO.
- AMLRO_interface: an earlier implementation of this GUI. amlro_gui is a rewrite of it, moving from server-rendered templates and session-cookie state to a Flask JSON API and a React/TypeScript frontend.
If you use AMLRO (the optimization engine this GUI wraps) in your work, please cite:
Kulathunga, D. P. et al. RxnRover/amlro. Computer Software. USDOE Office of Energy Efficiency and Renewable Energy (EERE), Advanced Materials & Manufacturing Technologies Office (AMMTO), 2026. DOI: 10.11578/dc.20260205.1
@misc{doecode_174798,
title = {RxnRover/amlro},
author = {Kulathunga, Dulitha Prasanna and Crandall, Zachery},
doi = {10.11578/dc.20260205.1},
url = {https://doi.org/10.11578/dc.20260205.1},
howpublished = {[Computer Software] \url{https://doi.org/10.11578/dc.20260205.1}},
year = {2026},
month = {feb}
}MIT License. Copyright 2026, Iowa State University. See
LICENSE for the full text, including U.S. Government rights
under contract DE-AC02-07CH11358 for Ames National Laboratory.
Developed with the assistance of Claude Code, Anthropic's AI coding assistant.