SUPERWRITERAI-powered writing. For everything you need to create.
Start with an idea
Bring a question, an outline, your data or an existing draft.
Write anything
Create papers, reports, slides, patents and grant proposals.
Make it yours
Review, refine and approve—with the final say on every change.
- Project-aware agents
- Skills & MCP
- Your models
- Your control
AURELIA_MAIN.TEX · OCT 18 AT 4:20 PM×Aurelia: Adaptive Coastal Resilience with a Multimodal AI Agent
\documentclass[11pt]{article}
\usepackage{amsmath, graphicx, booktabs}
\usepackage[numbers]{natbib}
\title{Aurelia: Adaptive Coastal Resilience with a Multimodal AI Agent}
\author{Lin Chen \and Mira Sol \and Daniel Park}
\begin{document}
\maketitle
\begin{abstract}
Coastal cities face compound risks from sea-level rise, storm surge and extreme heat. We present Aurelia, a multimodal agent that fuses satellite imagery, in-situ sensor streams and municipal planning documents to propose auditable adaptation strategies. Across four simulated cities, Aurelia raises intervention coverage while lowering projected cost and planning time.
\end{abstract}
\section{Introduction}
Adaptation planning is a decision problem under deep uncertainty. Planners must reconcile flood models, infrastructure inventories and community priorities that rarely share a common format~\cite{hallegatte2012,ipcc2022}. Existing decision-support tools either optimise a single hazard or require weeks of manual data preparation.
We argue that a planning assistant should (i) read heterogeneous evidence, (ii) reason over interacting hazards, and (iii) expose every recommendation together with its supporting sources. Aurelia is designed around these three requirements.
\section{Method}
\subsection{Multimodal evidence encoder}
Each city $c$ is described by observations $\mathcal{O}_c = \{x^{\mathrm{img}}, x^{\mathrm{sen}}, x^{\mathrm{doc}}\}$, mapped to a shared space by modality-specific encoders:
\begin{equation}
h_c = \sum_{m} \alpha_m \, f_m\!\left(x^{m}\right), \qquad \sum_m \alpha_m = 1 .
\end{equation}
\subsection{Strategy planner}
The planner proposes interventions $a \in \mathcal{A}$ and scores them by expected avoided loss, cost and equity impact. Every candidate keeps a trace to its evidence, so planners can audit and revise the plan.
\section{Results}
Table~\ref{tab:main} compares Aurelia with rule-based and single-hazard baselines across four simulated coastal cities.One platform for every document in a research career
Papers are just the start. The agent picks the matching Skills and templates for each document type and works with you from structure to wording — and anything missing can be added with Skills and MCP.
- Literature
- Data & code
- Your ideas
/super-brainstormCalls Skills and MCP tools per task- 01Research paper
/super-paperJournal & conference papers, project-level LaTeX - 02Literature review
/super-researchSearch, screen and synthesize into a structured review - 03Deep research report
/super-deep-researchMulti-round search with cross-checked sources - 04Editable slides
/super-pptGroup meetings, defenses and talks — exported as .pptx - 05Grant proposal
/super-grantRationale, research aims and technical roadmap - 06Patent
/super-patentInvention disclosure and application drafts - 07Software copyright
/super-copyrightManuals and filing materials - 08Defense materials
/super-defenseProposal, mid-term and thesis defense prep - 09Research promotion
/super-promoPlain-language summaries and outreach copy
Read papers with an AI guide
Reading is a daily research workspace, not a prelude to writing. Organize the library, read original and translated text side by side, and ask questions that point back to the source.
A library that remembers
Sort unread, in-progress and completed papers; track pages and reading progress.
Bilingual close reading
Follow the original PDF while reading a structured translation.
Ask with evidence
Ask what matters, receive a guided explanation, and trace every claim to its source marker.
Let the agent draft. You make the final call.
The agent reads the complete project and calls Skills to draft and revise. Every change arrives as a diff you can keep, reject or undo. Compile LaTeX locally and check source against the PDF.
Project-aware writing
Files, outline, editor and AI conversation share the same context.
You remain the reviewer
Accept or reject each proposed change; version snapshots make edits recoverable.
Paper-ready output
Compile with Tectonic and compare source with the finished PDF side by side.
Accept / Reject makes authorship visible and reversible.
The PDF utilities researchers actually need
Compress a submission that exceeds a portal limit, merge supplementary files, create a PDF, reorganize pages or convert formats. These are practical document tools — processed on your device.
- 01Compress and optimize
- 02Merge, split and organize pages
- 03Create, edit and annotate
- 04Convert to and from PDF
- 05Forms, privacy and repair
Local processing: useful documents do not need another cloud upload.
Extend the agent with Skills and MCP
SuperWriter is not a fixed feature set. Install or write Skills to teach the agent new research tasks; connect libraries, search, citations, your own tools and AI backends through MCP; reuse figures and research artifacts from the Gallery.
56 Skills
Ready-made Skills from brainstorming and peer review to defense and integrity checks — or write your own and hand a repeatable workflow to the agent.
Research connections
Connect Zotero libraries, files, search, citations and your own services as MCP tools the agent can call directly.
Scientific gallery
Organize figures, tables, citations, notebooks and generated visual artifacts by discipline.
A desk that looks like yours
Choose a dark research cockpit, a quiet light workspace or a warm typesetting theme.
Plus editor colour schemes: Paper Dark · One Dark · Dracula · Solarized Light · Rosé Pine Dawn · Cobalt · Ayu Light · Espresso and more
Data availability statement
Projects, papers and notes stay on your device. PDF utilities process files locally. SuperWriter handles licensing and activation; your research workspace remains yours.
AI use statement
- AI can be wrong. It may misread a paper, invent a citation, miscalculate or sound more certain than it should. Treat every output as a draft and check each reference and number.
- You stay accountable. Every change waits for your review; authorship and scholarly responsibility remain with you.
- Use it responsibly. Follow your journal's, institution's and funder's AI policies and disclose AI assistance where required. Never use it to fabricate data, plagiarise, or complete work that must be your own.
SuperWriter activation code
¥129one-time purchase
No monthly fee. No subscription.
- ✓ Every software feature included
- ✓ Free updates for future versions
- ✓ Bring Claude Code, Codex or your own API key
- ✓ One code for one device; reinstall on the same device anytime