magno

I built Wikli (https://www.wikli.com/) - an AI-powered news aggregator that clusters articles semantically and generates daily digests with editorial oversight.

The Problem: News fatigue is real. Reading 50+ articles daily (from hundred of different sources) to stay informed is unsustainable, but traditional aggregators just dump links without context.

Wikli uses a three-stage pipeline:

Scraping & Processing (Cloudflare Workers): RSS feeds → content extraction → AI classification Semantic Clustering (Python): Claude groups related articles across sources into coherent stories Digest Generation: AI synthesizes clusters into readable reports with context and TLDR

Technical Highlights:

Cloudflare Workers + PostgreSQL for scraping infrastructure Hybrid content extraction (Readability + Puppeteer fallback for tricky sites) Claude Sonnet 4 for clustering and synthesis (outperformed embedding-based approaches) Theme-based filtering with relevance scoring (0-10 scale per article) Telegram bot with stateless approval workflow for editorial control

What's Different:

Semantic clustering beats chronological or source-based grouping Context from previous digests prevents repetition Human-in-the-loop via Telegram for quality control (can edit title/approve digest) Open architecture: separate Brief Generator (Python) and Scraper API (TypeScript)

Stack: TypeScript, Python, PostgreSQL, Drizzle ORM, Claude/Gemini APIs

The system handles rate limiting across domains, AI API throttling, and includes a DataManager abstraction for centralized data operations. Currently live in Italian at wikli.com - language-agnostic by design but focused on the Italian market for now. A the moment running with two topics (AI innovation and Inter Milano Football Club) via Telegram and wikli.com website.

Happy to get any feedback.

I started http://blogo.it as a side project. It's now the third media property in italy with 11 ml/uniques month. Before Blogo I started another company as a side project, it was a web agency I later sold to a pre ipo company. Starting a project in this way is a really good way to start slowly, think about it a lot and to have no pressure at the beginning.

You are right, tipsandtrip is a relatively simple project (web + api + iphone) but I think that a similar process can be used also on larger project to build a MVP that can be launched and shown to investors. Better to have something working to show than a couple dozen of slides ;-)

No revenues at the moment. Working with remote teams is fine if you use the right tools (redmine, svn for example) and if you try to avoid working with people on the other part of the ocean. We worked with people in Argentina, Japan, India, Ukraine and Russia. The last two are at the moment the countries where we find the highest programming skills.

We will try first to build a community/traffic and then will look for revenues. Revenues will be adv based (display, affiliation with travel sites).

@fabio, yes, exactly what we did. We gave it a shot and wanted also to test the many different services available to crowdsource workforce.

Here to answer any question you have on tipsandtrip :)