StoryTrace - Git for News

StoryTrace is a "Git for News" — a tool that tracks how a story mutates as it travels across global media outlets.

You paste any article URL or type a topic (e.g. "Iran nuclear deal"). StoryTrace's 7-agent LangGraph pipeline springs into action: the Seed Agent finds the original wire story via GDELT, the Crawler Agent scrapes 15 RSS feeds across 8 countries, a Translator Agent localizes non-English coverage to English, the DNA Extractor (Featherless/Qwen2.5) pulls structured facts from each article, the Drift Scorer measures how many key facts each outlet dropped or distorted, the Geo Builder organizes coverage by country into a D3-ready tree, and an Alert Agent fires webhooks when drift exceeds a threshold.

The results render as an interactive D3.js tree — nodes colored green (faithful), amber (moderate drift), red (high drift) — alongside a 3D globe showing which countries covered the story and how differently. Clicking any node reveals a DiffPanel showing exactly which facts were kept, dropped, or altered.

The backend is FastAPI + LangGraph + PostgreSQL + Redis. The frontend is Next.js 16 with D3 v7 and react-globe.gl. AI layers: Featherless API (Qwen2.5-7B) for structured JSON fact extraction, Google Gemini for translation and optional world-impact forecasting, spaCy for local NER (zero tokens). The entire pipeline runs on ~4,000–6,000 tokens per story — lean by design.

StoryTrace addresses a real problem: media consumers have no tool to see how coverage diverges from the original facts. We make that drift visible, quantified, and interactive.

Note: the demo URL is substituted with the GitHub repository link.