AI/ML●●Solid
Gemma 4 based local RAG on 25 Years of news articles
500k-article Swiss Teletext corpus makes this RAG demo actually interesting.
Niche GemBig Brain
folli
102mo ago
Back to browse
GitHub Repository
A fully local Retrieval-Augmented Generation (RAG) implementation for querying 25 years of Swiss Teletext news (500k articles in German language)
11 starsJupyter Notebook
Local RAG on 25 Years of Teletext News
by folli·Apr 1, 2026·2 points·0 comments
AI Analysis
●●SolidNiche Gem
Local RAG on 500k teletext articles when most demos use toy datasets.
Strengths
- •500k document corpus is genuinely non-trivial compared to typical RAG demos
- •Fully local execution with no APIs keeps sensitive data on your machine
- •Hybrid search combining vector and full-text improves retrieval accuracy
Weaknesses
- •German-language corpus limits applicability for most English-speaking developers
- •Standard RAG architecture without novel technical approaches beyond dataset choice
Category
Target Audience
Developers building local RAG pipelines, privacy-conscious users
Similar To
PrivateGPT · LlamaIndex · LangChain
Post Description
A fully local Retrieval-Augmented Generation (RAG) implementation for querying 25 years of Swiss Teletext news (~500k articles in German language) — no APIs, no data leaving your machine.
Why? I thought it's a cool type of dataset (short/high density news summaries) to test some local RAG approaches.
Similar Projects
Education○Pass
Tracking Top US Science Olympiad Alumni over Last 25 Years
Just a Perplexity query result, not a tool you can actually use.
Niche Gem
bkls
601mo ago
AI/ML●●Solid
Local Context and Memory Stack
Tops LongMemEval and LoCoMo benchmarks with local-first AI memory architecture.
Big BrainShip It
dhravya
103d ago
Developer Tools●●Solid
Code-Graph-RAG – Knowledge graph RAG for any codebase
Knowledge graph RAG for code when Cursor and Cody already own this space.
Big BrainSolve My Problem
vitali87
113mo ago
AI/ML●●Solid
A 150M model that extracts verbatim evidence spans for RAG, no LLM call
150M model replaces LLM calls for evidence extraction with comparable F1 scores.
Niche GemBig BrainSolve My Problem
justacoolname
603d ago
AI/ML●Mid
Query years of Ask HN and Show HN discussions as a knowledge graph
Single-file DB for LightRAG is clever, but HN scraping is solved by existing tools.
Ship ItNiche Gem
swaminarayan
451mo ago