For researchers · MES scientists
Design your next experiment from the whole field — not the last five papers you read.
The evidence you need is trapped across tens of thousands of PDFs, in inconsistent units, impossible to compute against. MESSAI turns it into one structured, predictive instrument — so you decide from the corpus, not from memory.
23,596 papers · 202K+ measurements · 835 parameters · free for early-career researchers
What you get
Three things every MES researcher is actually trying to do.
01 · Find what matters
Find what’s known and relevant — fast.
Search two decades of results as one structured corpus, follow the causal edges between parameters, and get papers, gaps, and outliers ranked to your system class — instead of deciding from the last five papers you happened to read.
Open Research Intelligence02 · Design before the bench
Design and simulate before touching a bench.
Build a reactor in an interactive 3D lab, sweep the inputs that matter, and watch predicted polarization, power, and scale-up respond in real time — so costly trial-and-error happens in the browser, not the fume hood.
Open the 3D Laboratory03 · Trust the number
Trust every number, with calibrated uncertainty.
Power density for ostensibly identical cells swings by orders of magnitude — corpus-wide CoV is roughly 1,285%. Every prediction ships as a value, a unit, a 95% calibrated interval, a confidence, and a traceable source. Never a naked point estimate.
See PredictionsCorpus CoV ≈ 1,285% — a trustworthy interval for a novel design often spans an order of magnitude. We show the range, not a fiction.
The instruments
Five surfaces, each pointed at one of those jobs.
Serves outcome 01
Knowledge graph
835 parameters wired by 2,819 causal couplings, so you trace what drives a metric rather than reading its distribution in isolation.
Open
Serves outcome 01
Personalized recommendations
Papers, parameters, and open gaps ranked to your system class and reading history — the relevant handful surfaced from 23,596.
Open
Serves outcome 02
The Lab + AI copilot
Design a reactor in interactive 3D and ask the copilot to sweep inputs and predict performance before you commit a single electrode.
Open
Serves outcome 03
Reproducibility scorer
Score any paper on methods completeness, so you know whether a result is solid enough to build on before you cite it. Scoring runs on demand — most of the corpus is not scored yet.
Open
Serves outcome 03
Anomaly scanner
Measurements that sit far outside the corpus distribution are flagged automatically, so a mis-normalized or fabricated number never enters your analysis unseen.
Open
One login. Every instrument reads from the same structured corpus, so a paper you scored, a parameter you swept, and a prediction you trusted all point at the same evidence.
The corpus behind every answer
Not a demo dataset — the whole field, structured.
Spanning 17 MES system types · 18 mapped research gaps · calibrated to a 95% interval (power-density CoV ≈ 1,285%, conformal-corrected)
Where the data comes from
Every number traces back to a published, peer-reviewed source.
Source
Peer-reviewed literature — 23,596 papers across hundreds of journals and two decades of MES research.
Acquisition
Pulled through open, legitimate channels — Unpaywall, arXiv, Europe PMC, Semantic Scholar, CrossRef, OpenAlex. No Sci-Hub, no paywalled scraping.
Provenance
Every value is carried with its source paper, units, conditions, confidence — harmonized to SI so studies finally compare like-for-like.
Platform figures as of August 2026 · acquisition chain: Unpaywall → arXiv → Europe PMC → Semantic Scholar → CrossRef → OpenAlex
Get started
Put the whole field behind your next result.
Free for early-career researchers. Request access and start designing from evidence instead of memory today.