cat ~/projects/babelbit/README.md

babelbit.net

data platform · NLP · international media · 2026

A geopolitical narrative comparator: it gathers news from 100+ outlets in 25 countries, groups the pieces that cover the same event and shows side by side how each one tells it.

visit site ↗

## the problem

The same event reads very differently in Moscow, Washington, Beijing or Buenos Aires, and comparing them means reading in several languages. BabelBit does not decide what is true: it shows each outlet's framing, tone, coverage and declared bias, and records who predicted what.

## what it does

Clustering of articles in 8 languages by semantic similarity (multilingual embeddings + pgvector)
Per-event comparison: tone, keywords, who broke it first and a coverage tree
Economy, Outlook and Technology sections with data from the IMF, FRED, Polymarket and OpenAlex
Source predictions extracted with Claude, reviewed by a person and tracked on a scorecard
D3 data maps: outlet network, cross-region coverage and actors linked to Wikidata
Automatic posting to Facebook, Instagram, Bluesky and Telegram with generated cards

## stack

.NET 10 · Clean Architecture
ASP.NET Core
EF Core
PostgreSQL + pgvector
Python · FastAPI
Transformers (NLP)
Claude API
Angular 22 SSR
Tailwind 4
D3
Linode · nginx · systemd

## notes

Built with AI-assisted engineering (Claude Code): a monorepo with 18 documented architecture decisions (ADRs) and 54,000+ articles analysed. In production since October 2026 on four Linode servers linked by a private network.

something similar? let’s talk