Nexus
Nexus is a full-stack platform that transforms how scientists run experimental optimization campaigns.
Core trick
Wraps Bayesian optimization suggestions with LLM-written explanations and detects instrument drift in experimental campaign data.
Tags
Description
Nexus is a full-stack platform that transforms how scientists run experimental optimization campaigns. Instead of manually choosing algorithms, tuning hyperparameters, and guessing when to switch strategies, Nexus autonomously diagnoses, decides, and explains — acting as a scientific reasoning partner, not a black box.
Scientists spend 80% of optimization time on algorithm selection and debugging, not science. Existing tools like Optuna or Ax give you the next point to evaluate — but they don't tell you *why*, they can't detect when your instrument has drifted, and they start from scratch every campaign. Nexus tells you WHY, not just WHAT's next.
README excerpt
# Nexus **Intelligent Optimization Platform for Scientific Experiments** 🌐 **[View Landing Page](https://sissifeng.github.io/HELIOS_Nexus)** — Interactive product showcase --- **Intelligent Optimization Platform for Scientific Experiments** Nexus transforms how scientists run Bayesian optimization campaigns — from black-box tool to transparent research partner. Upload your data, get intelligent suggestions, and understand every decision the optimizer makes. --- ## The Problem Scientists running experimental optimization (materials discovery, drug formulation, process engineering) face two barriers: - **Opacity** — Existing tools don't explain why a suggestion was made - **Programming requirements** — Most tools demand ML expertise to configure The result: researchers don't trust suggestions, can't diagnose stalled optimizations, or abandon the tool entirely. ## The Solution Nexus wraps Bayesian optimization with a diagnostic intelligence layer: 1. **Zero-code setup** — Upload CSV, map columns visually, start optimizing 2. **148+ inline visualizations** — Real-time diagnostics that surface problems before they waste trials 3. **Transparent suggestions** — Every recommendation includes novelty scores, risk profiles, and provenance 4. **AI chat** — Ask "Why did you switch strategies?" and get answers backed by computed signals 5. **Campaign memory** — Decision journals, learning curves, and hypothesis tracking --- ## 🚀 Quick Start (4 Options) Choose the deploymen