Planet Earth Wrapped

Create a list of all major nations. On a running weekly basis, spin up an LLM for each nation to search for all major good & bad news to come out of that nation over the past 7 days and store in a structured format with supporting images and other references. In the final week of every year, produce a nice slideshow on a public url that summarises the year across all nations, with thematic highlights like 'Most heroic individual', 'Saddest day of the year', 'Celebrities', 'Inventions', and many others, rotated based on people's engagement. Show the good, bad, and the ugly of all of humanity.

Spotify Wrapped, Russell Howard's Good News, YouTube Recap, and the BBC, all do one thing everybody loves: tell us what noteworthy events happened recently. But they are limited to only having enough bandwidth and media attention to focus on local events. AI could systematically round up major events from every single place on planet Earth and update an annual page with the results in an entertaining, inspiring, and saddening way. There has never been so much rapid change in human history. Every year has defining moments which are quickly replaced by the next year's. It would be nice to have a mutual way to reminisce on the past year's events.

Produce a full product requirements document defining the MVP and long-term vision, target users, yearly and weekly workflows, definition of 'major nations', event inclusion criteria, good/bad/ugly taxonomy, editorial standards, citation requirements, risk register, and a staged technical architecture for building Planet Earth Wrapped responsibly.

Build the canonical country and region registry using ISO-style identifiers, aliases, demonyms, languages, time zones, regions, disputed-territory handling notes, and initial country metadata. Include seed data, validation utilities, tests, and documentation for how the rest of the system references nations consistently.

Design and implement the core schema for weekly events, countries, sources, citations, media assets, themes, scores, categories, people, organizations, locations, translations, editorial status, and yearly recap groupings. Produce database migrations, TypeScript/Python models, validation rules, sample records, and schema documentation.

Create a source discovery framework for RSS feeds, public news APIs, reputable international outlets, local outlets, government updates, NGO reports, and fact-checking sources. Include source quality scoring, language coverage strategy, per-country seed source lists, robots/copyright considerations, and code for managing and auditing source metadata.

Implement a first working ingestion pipeline that can collect recent articles from configured RSS/API/search-style connectors, normalize metadata, deduplicate URLs, cache raw source records, handle failures, and store candidate news items for a small pilot set of countries. Include tests, fixtures, and local run instructions.

Build the LLM prompt and orchestration layer that converts raw articles into structured candidate events with title, summary, country, date, people, organizations, sentiment, importance, category, citations, uncertainty, and quote/source references. Include JSON schemas, retry logic, validation, multilingual handling notes, and test fixtures.

Add translation and localization support so non-English source material can be summarized in English while preserving original-language citations and key names. Include language detection, translation abstraction interfaces, localized display fields, transliteration handling, tests, and documentation for extending to more languages.

Implement logic to group multiple articles into the same real-world event, merge citations, detect duplicates across countries and languages, rank importance, estimate confidence, and balance national versus global salience. Include scoring algorithms, explainable ranking fields, test datasets, and evaluation utilities.

Create the classification system for positive, negative, tragic, inspiring, controversial, scientific, cultural, sports, political, environmental, humanitarian, and other event dimensions. Include taxonomy docs, LLM classification prompts, calibration examples, edge-case handling, bias warnings, and tests for category consistency.

Build a media metadata layer for supporting images and references, emphasizing licensed, embeddable, or attributed media rather than unsafe scraping. Include image URL storage, attribution fields, license labels, fallback visuals, thumbnail metadata, citation rendering helpers, validation scripts, and documentation on legal-safe media use.

Implement an admin workflow for reviewing candidate events before public display, including approve/reject/edit states, confidence warnings, citation inspection, sensitive-content flags, correction notes, audit history, and reviewer assignment. Provide backend routes, frontend admin screens, mock auth, tests, and editorial runbook documentation.

Build the public-facing weekly explorer where users can browse countries, regions, event categories, good/bad/ugly filters, timelines, citations, and media cards. Include responsive frontend pages, API endpoints, loading/error states, accessibility basics, sample data, shareable URLs, and end-to-end tests.

Create privacy-conscious analytics for views, shares, category clicks, country interest, and slide engagement. Implement a theme rotation engine that uses engagement signals to prioritize recap categories while avoiding over-optimization toward sensationalism. Include event tracking interfaces, aggregation jobs, admin settings, and tests.

Build the system that transforms approved weekly events into end-of-year recap candidates such as 'Most heroic individual', 'Saddest day of the year', 'Major inventions', 'Celebrity moments', 'Scientific breakthroughs', 'Global crises', and regional highlights. Include narrative templates, ranking logic, citation requirements, and sample generated recap data.

Implement the public annual slideshow at stable URLs, with responsive slides, animations, thematic sections, country and region highlights, citations, image attribution, share cards, accessibility support, SEO metadata, and graceful fallbacks for missing media. Include demo data and frontend tests.

Create deployment-ready infrastructure code and operational scripts for weekly ingestion, daily processing retries, annual final-week recap generation, background queues, environment configuration, logging, monitoring hooks, rate-limit controls, and local/staging/production setup documentation.

Develop evaluation tools for factual consistency, citation coverage, source diversity, geographic balance, duplicate detection, ranking fairness, translation quality, and hallucination prevention. Include benchmark fixtures, automated checks, manual QA checklists, reporting dashboards or scripts, and improvement recommendations.

Produce and implement safeguards for copyright-safe media use, source attribution, defamation risk, sensitive tragedy coverage, public-person versus private-person handling, privacy-conscious analytics, takedown/correction workflows, and content moderation policies. Include code-level guards, policy documents, and reviewer guidance.

Integrate the backend, ingestion pipeline, LLM extraction interfaces, review workflow, weekly explorer, analytics, and annual slideshow into a coherent MVP using seeded sample data and a small pilot country set. Include a complete local demo script, test run, known limitations, and deployment checklist.

Prepare the project for a limited beta by polishing UX, improving error handling, adding onboarding/admin documentation, writing weekly operations playbooks, creating incident response procedures, defining KPIs, and documenting the roadmap from pilot coverage to all countries.