Ajako Taja
AI-powered news platform that monitors hundreds of sources, scores stories for real multi-source traction, and delivers concise, continuously updated summaries.
- Role
- Creator
- Type
- Personal Project
Overview
A personal project I built while teaching myself AI. Ajako Taja watches hundreds of RSS feeds and scores each story on source count, coherence, recency, and named-entity agreement, then has an LLM summarize only the clustered source material — no content beyond what's in the sources. Readers pick the niches they care about (tech, crypto, sports, politics, entertainment, and more) and get a personalized feed instead of one firehose. Built with Next.js. Not affiliated with any employer.
The problem
News is scattered across hundreds of outlets, and a single loud headline is easy to mistake for a real story. Ajako Taja looks for stories that several independent sources are actually covering, and condenses them into short, sourced summaries.
What I built
- Designed and built the product end to end — public reader site, staff dashboard, and the content pipeline.
- Built the story-grouping and scoring logic that clusters items covering the same real-world story.
- Kept the LLM step grounded: it only summarizes the clustered source material.
Architecture & key decisions
- Two Next.js apps on Vercel: the public reader site and a staff dashboard that also hosts every scheduled job.
- Vercel Cron pipeline: hourly feed ingestion, periodic story grouping, AI article drafting, image generation, and social distribution.
- Supabase Postgres (with pgvector) for data and Supabase Storage for images.
- Editors can review each pipeline step, and per-niche automation toggles control how far a story goes without human approval.