Lumioly

AI-powered news and tools intelligence platform — live articles, curated AI tools, and Gemini-powered "Why This Matters" analysis in one place.

Comment

The Problem

The AI space moves faster than anyone can keep up with. Every day brings new models, new tools, and breakthroughs — and an overwhelming flood of articles to match. Developers in emerging markets like South Africa have it worse: expensive newsletters, paywalled content, and no centralised place to track what actually matters.

The Solution

Lumioly is a fully autonomous AI intelligence platform. It pulls live news from five major tech sources in real time, curates 44 AI tools across six categories with on-demand Gemini explanations, and features a "Why This Matters" analysis — where Gemini reads each article and generates a three-sentence breakdown of its real-world significance for developers.

What I Built During DSOC 2026

  • Lumioly — designed and built the full platform from my own design framework, including live RSS news aggregation from five major tech sources, the AI tools directory, Gemini-powered chat, and the dark luxury aesthetic across desktop and mobile
  • Why This Matters — the centrepiece feature: Gemini reads every article and generates a three-sentence analysis explaining its real-world significance to developers right now
  • Topic restriction & prompt injection protection — AI system prompt that keeps Lumioly strictly focused on AI and tech, with graceful redirects for off-topic queries
  • QA & edge case testing — boundary testing including off-topic queries, borderline questions, and prompt injection attempts
  • UX/UI Design — full Figma documentation with desktop and mobile screens, design system, and clickable prototype
  • Full deployment — live on Railway with PostgreSQL, WhiteNoise, and autonomous RSS pipeline

How I Built It

Django 5.2 backend with Python 3.13 handles all routing and data logic. Google Gemini 2.5 Flash powers the AI chat, per-tool explanations, and article analysis. RSS feeds via feedparser pull live news automatically from TechCrunch, VentureBeat, The Verge, Ars Technica, and Wired on every page load — no database, no manual curation, always fresh. YouTube Data API v3 fetches relevant tutorials for each tool automatically. The entire platform is deployed on Railway with PostgreSQL and runs autonomously.

Challenges

Google deprecated the entire google-generativeai SDK mid-build with almost no warning, requiring a full migration to the new google-genai package. Passing article data safely through URL parameters without breaking the HTML also required replacing inline Django template tags with a JavaScript function — a subtle but critical fix.

What I Learned

  • Never pin to older AI SDKs — deprecations are real and painful
  • Autonomy is a feature, not an afterthought
  • Prompt injection protection should be built in from day one, not added later
  • Standing out visually is a strategic decision — the dark luxury aesthetic was intentional and every piece of feedback confirmed it was the right call
  • Designing for real constraints (emerging markets, data access) produces better products than designing for ideal conditions

Built With

Share this project:

Updates