Personalized AI & Technology Intelligencein-progress

KnowledgeHub

Personalized AI & technology intelligence, delivered weekly.

Next.js 16React 19TypeScriptTailwind CSS 4shadcn/uiPrismaPostgreSQLRedisPythonOpenAIResendTurborepoDocker

Intelligence Pipeline

KnowledgeHub

Interactive preview — try the live demo for the full experience

7Workspaces

11Topic categories

5 seededRSS sources

5Ranking signals

Overview

KnowledgeHub is a personalized technology intelligence platform — not simply a newsletter. A Python collector worker continuously ingests articles from curated RSS sources and scrapers, enriches them with AI (summaries, topics, keywords, embeddings, quality scores), and stores them in a shared PostgreSQL database. A composite recommendation engine ranks content per user from topic weights, engagement, freshness, and diversity — without relying solely on LLM prompts. Users receive beautifully formatted, tracked weekly editions via email.

Highlights

Python collector worker decoupled from Next.js, sharing a Prisma-managed PostgreSQL schema

5-signal composite recommendation engine (quality, affinity, engagement, freshness, diversity) with Redis caching

Offline-capable ingestion pipeline with OpenAI enrichment and heuristic fallbacks when keys are absent

Full admin dashboard for sources, articles, worker runs, newsletters, and analytics

Features

›SaaS marketing landing with sample newsletter preview and onboarding flow
›Magic-link authentication and weighted topic preferences across 11 categories
›RSS and scraper collection with AI enrichment, embeddings, and global quality scoring
›Personalized newsletter generation with HTML rendering and click/open tracking
›Composite ranking with greedy diversity selection and 5-minute Redis cache TTL
›Admin console — users, news sources, articles, newsletter editions, worker executions
›Domain-driven modules with typed domain events for future ML and search features
›Docker Compose local stack with PostgreSQL 16 and Redis 7
›Graceful degradation — Redis, OpenAI, and Resend are optional for local development
›Turborepo monorepo with shared scoring primitives across TypeScript and Python

Technical

Built as a Turborepo monorepo: `apps/web` (Next.js 16 App Router with domain modules for auth, preferences, recommendation, newsletter, analytics, admin), `workers/collector` (Python feedparser + BeautifulSoup pipeline writing via SQLAlchemy to the same Prisma schema), and packages for database, shared scoring/topics/events, and shadcn/ui. The recommendation engine blends five weighted signals in TypeScript; the collector enriches and embeds offline when OpenAI is unavailable. Email delivery uses Resend with open-pixel and click tracking feeding engagement signals back into ranking.

Learnings

01Composite scoring beats prompt-only personalization for explainability, cost control, and tunable ranking
02Optional Redis, OpenAI, and Resend degrade gracefully — the full stack runs locally without external services
03Shared scoring primitives in `@workspace/shared` keep the web app and Python worker conceptually aligned

All personal projects