WorldLens

Inspiration

A headline can tell us that something happened without helping us understand the place, the history behind it, or the people reporting it. When the story is a humanitarian crisis, the question often becomes: what can I actually do?

We built WorldLens to connect those questions in one place. It is a 3D world-intelligence globe where you can explore today’s news, move through history since 1800, examine the sources, and review a possible next step. Our approach is simple: understand, verify, then act.

WorldLens is a Noema product for students and teachers, curious news readers, and researchers who need geographical and historical context.

What it does

Explore 1800 to LIVE. Rotate a realistic CesiumJS globe, search for a place or event, and move through time from centuries down to hours, within the precision supported by the data.
See the world of the selected date. Explore dated historical borders and flags, historical events and battles. A selected country can follow its changing borders and recorded predecessors or successors as you move through time. Historical coverage is incomplete, and approximate data is identified.
Read stories in context. Live news from GDELT is grouped into events with attached articles rather than placing every article on its own pin. Wikipedia-based past-news search extends exploration back to 2000, alongside the curated historical dataset.
Inspect the reporting. Event cards show the original sources, publisher types, ownership where known, and where outlets are based. Framing comparisons distinguish facts, claims, allegations and assessments using the available headlines and snippets. Political ratings describe publishers only and include provenance; most outlets remain unrated.
Ask an AI guide. In Agent mode, plain-language requests can move the camera, change the date, open events, highlight places, toggle layers or create a historical tour. Chat mode answers without operating the globe. Signed-in users can save chats and control whether their previous chats and viewing history inform the assistant.
Review humanitarian action. On a clearly labelled humanitarian demo scenario, the assistant can find an action option and open its review screen. The person checks the recipient, purpose and amount, then approves in PayPal themselves.

The charity feature is still in testing. All checkout activity uses PayPal Sandbox, no real money moves, and the test recipient is a demo merchant rather than a live charity. Informational links to humanitarian organisations are separate from the checkout and do not imply affiliation.

How we built it

WorldLens is a TypeScript monorepo. The frontend uses React, Vite, CesiumJS, Tailwind CSS, Zustand and TanStack Query. The API uses Hono and Zod, with shared schemas for events, temporal data and AI tools. PostgreSQL with PostGIS on Neon stores live news, accounts and payment records.

The hosted version serves the frontend through Cloudflare Workers and runs the Node.js API on Render. A same-origin proxy connects the two so sessions and API requests work under one website address. Render is an additional hackathon sponsor tool used in the deployed application.

Historical geography combines OpenHistoricalMap, Natural Earth and Wikidata. Wikipedia supplies linked context and past-news records. The news pipeline normalises and deduplicates articles, resolves locations, clusters related coverage and preserves story identity across refreshes. The repository also includes an optional RSS provider.

AI in the product: the deployed assistant uses Google Gemini. A bounded tool loop validates model requests against shared Zod schemas, and the browser validates actions again before applying them. AI also supports short news summaries and source-framing comparisons grounded in the supplied headlines and snippets. The provider layer additionally supports Anthropic and OpenAI-compatible services. Rate limits, usage accounting, caching and a daily AI budget control consumption, with a rule-based offline assistant available as a fallback.

AI used during development: we used Claude Code with Claude Opus 5.5 to develop WorldLens. This development assistance is distinct from the Gemini assistant running in the hosted product.

PayPal: the browser uses the official PayPal JavaScript SDK v6 for eligibility checks and checkout sessions. The backend uses OAuth 2.0 and Orders API v2 to create, retrieve and capture sandbox orders. It validates the reviewed terms and amount, keeps the client secret on the server, and uses idempotency keys. A receipt depends on the server-confirmed PayPal capture result. The code also implements signature-verified webhook reconciliation, duplicate-delivery protection and guarded payment-state transitions; webhooks are not yet configured on the hosted demo.

The AI has no tool that can create, approve or capture a payment. It can prepare the review experience, while the human controls checkout.

Challenges we ran into

Historical maps require more than changing a year label. We had to account for dates of validity, incomplete coverage, coastlines versus land borders, dated flags and country succession without presenting every reconstruction as exact.

News coverage also changes between refreshes. Deduplication and cluster continuity help keep one developing story together while preserving its growing set of articles and sources.

Another challenge was connecting model-generated instructions to a working interface. We used a small, explicit tool surface, validation on both sides of the network, and separate Chat and Agent modes. Payments remain outside that tool surface.

Rendering a detailed globe alongside labels and panels required screen-space clustering, horizon culling, viewport-based requests and adaptive quality. The performance manager adjusts resolution and detail from measured frame times. We do not claim a universal frame rate across devices.

Accomplishments we are proud of

We connected live news, historical geography, source inspection and a human-approved PayPal sandbox flow in one working website. The interface includes a text events list, keyboard navigation, reduced-motion support and responsive layouts alongside the globe.

The repository includes Vitest coverage for shared schemas, AI actions, news processing and payment behaviour, plus Playwright application and responsive tests and database checks. Payment end-to-end tests use test doubles, so we distinguish automated coverage from live sandbox verification.

This is a new project built during the hackathon. The repository began on 14 October 2026 with a placeholder README and no application code. The application, data pipelines, PayPal and AI integrations, deployment and documentation were developed during the submission period.

What we learned

We learned how to turn an AI answer into a useful, validated sequence of interface actions, and how to keep uncertainty visible when combining historical and news datasets. Building the PayPal flow also reinforced the importance of explicit human review and server-confirmed payment status. Keeping the demo usable through provider failures and limited hardware became part of the product design.

What’s next for WorldLens

We plan to test with students, teachers and news readers, improve historical and news coverage, and measure performance and operating costs on everyday devices. The charity feature will remain in sandbox testing until the recipient arrangements and production flow are ready.

The business plan is free exploration with possible Pro, Classroom and Teams offerings. Alerts, shareable views, classroom tours and exports are planned features, with pricing still to be tested.

Team

Hasan Abdullah, founder of Noema and team lead, built the first version and leads globe development, historical geography and optimisation. Evan Setyawanta Sitepu develops the PayPal and Take Action checkout. Mohammad Nabil Hanif develops live news and the interface. All three studied at President University in Indonesia.

Try WorldLens

Built With

cesiumjs
claude-code
cloudflare-workers
gemini
hono
neon
node.js
paypal
postgis
postgresql
react
render
tailwindcss
typescript
vite
zod
zustand

Submitted to

PayPal AI Hackathon

Created by

What is my contribution. I redesigned the interface with a glass-style visual treatment, giving the product its monochrome look, a top navigation dock, a detached context card, simplified command bar controls, and a bottom time dock. The purpose was to make the map experience feel modern and unobtrusive, keeping attention on the globe while the controls stay close at hand. Within that redesign I fixed the clock display, made the context card toggleable, and kept the interface legible over bright satellite imagery.
I also reworked the side panels so that the help panel and layer controls open beside the context card instead of covering the screen. This lets users read information and change settings at the same time, without losing sight of the map or closing one panel to use another. I refined the spacing, dividers, and toggle states in those panels and improved the search and text input alignment so the controls feel balanced and easy to read.
On the feature side, I added support for an OpenAI-compatible AI provider so the assistant can run on different models, introduced RSS as a second news source alongside GDELT so coverage is not dependent on a single feed, and built the assistant chat history with the ability to pin, rename, and delete conversations. These exist so users can choose their AI backend, receive broader and more reliable news coverage, and keep track of past conversations instead of losing them. I also added a small tooling improvement that keeps the project's formatting checks passing and prevents installed skill assets from cluttering the project, which helps the team maintain a clean and consistent codebase.

Nabil Hanif
Team lead and main developer. I created the repository, set up the architecture and built most of WorldLens, working with Claude Code as my coding agent, and I merged my teammates' branches into the main build.

Foundation: scaffolded the monorepo: the React and CesiumJS web app, the Hono API, the shared temporal model and Zod schemas, the AI assistant's validated tool layer, the first PayPal sandbox checkout (Evan then updated it), and PostgreSQL/PostGIS on Neon.

History on the globe: day-precise borders from 1800 to today from OpenHistoricalMap (about 2,400 dated countries, including occupations and puppet states), 5,102 battles from Wikidata, the flag each country used on the selected date, territorial succession, and a realistic night side with city lights.

Timeline and map: reworked the time navigator (drag, zoom, play, typed dates), flicker-free event clusters, a picker for several events at one place, place names on the globe, and the phone layout.

News: live news from GDELT's 15-minute files, one event per story across refreshes, fixes to how headlines are pinned to places, past news back to 2000 from Wikipedia's Current events, and publisher details from Wikidata.

AI assistant: Google Gemini support so it runs on a free tier, opt-in memory of a user's history and past chats, and clickable links in answers.

Accounts: sign-in with Google or email (verified by an emailed code), with chats and viewing history saved per account.

Shipping: deployment on Cloudflare Workers, Render and Neon, and tests alongside each feature (the suite now has about 400 unit tests plus Playwright end-to-end runs).

In progress on its own branch: day-by-day front lines for 38 theatres of war (1812 to 1995), still being checked for accuracy.

Hasan Abdullah
Contributed to the WorldLens AI project by integrating PayPal payment functionality and handling application deployment. Responsible for connecting the PayPal payment system to support transaction processing and ensuring the application was successfully deployed and accessible online.

Evan Setyawanta Sitepu