Project
Swarm Signal
A proof of concept for autonomous content: an AI agent system that reads arXiv research papers most people would never find, analyses them, and writes them up as genuinely useful articles for an AI-literate audience. Over 100 signals published and counting.
The point isn't just the articles - it's showing that an agent pipeline can produce content worth reading, not just content that exists.
Why I built it
There are thousands of AI research papers published every week. Most of them never get read outside academia. Meanwhile, the AI coverage that does reach people is either hype or surface-level summaries. I wanted to see if an agent system could bridge that gap - automatically finding papers that matter, extracting the key insights, and writing them up in a way that's actually useful.
Swarm Signal is the result. The whole pipeline is autonomous: topic discovery, paper analysis, article generation, audio narration, and publishing to Ghost CMS. I set the editorial direction and quality bar, but the agents do the work. It's a live proof of concept for what AI content pipelines can look like when you take quality seriously:
- Sources from primary materials - arXiv papers and technical documentation, not press coverage or Twitter threads.
- Takes positions - evidence-based analysis rather than hedged summaries that say nothing.
- 100+ articles published - covering agents, swarms, reasoning, safety, and real-world deployment.
- Built for an AI-literate audience - people who build systems, not people who just follow the news.
Six research verticals
Content is organised into focused research areas, each covering a distinct aspect of the AI landscape:
Agent Design
Architectures, tool use, and frameworks for building AI agents. From single-agent patterns to production deployment strategies.
Swarm Systems
Multi-agent coordination, swarm intelligence, and collective behaviour. How groups of AI agents work together - and when they don't.
Reasoning & Memory
How models think, remember, and build context. Chain-of-thought, retrieval-augmented generation, and long-term memory architectures.
Safety & Governance
Alignment, oversight, and the policy landscape. What matters for building responsibly, beyond the talking points.
Models & Frontiers
Frontier model developments, benchmarks, and capability analysis. What's real, what's noise, and what it means for builders.
Real-World AI
Deployment, robotics, embodied AI, and practical implementation. Where theory meets production and the friction nobody planned for.
How it works
Signals
Shorter, timely research analysis pieces. Typically 5-6 minute reads covering a specific development, paper, or trend with context and a clear takeaway.
Guides
Longer, foundational deep-dives. 10-13+ minute reads that explain a concept thoroughly with practical implications for builders and decision-makers.
Audio
Every article is available as audio for listening on the go. Queue articles and listen through the site's built-in player.
The pipeline
The entire content operation is autonomous. Agents scan for research themes, pull papers from arXiv, analyse them, write structured articles, generate audio narration, and publish to Ghost CMS - end to end. I define the editorial standards and quality bar, but the pipeline runs itself.
It's a deliberate proof of concept: can an AI agent system produce content that's genuinely worth reading? After 100+ published articles, the answer is yes - if you design the system properly and set the bar high enough.
Built with
| CMS | Ghost (self-hosted) |
| Hosting | Docker on Hetzner VPS (same platform as tylerbuilds.com) |
| API | Custom Node.js API for audio, search, and content operations |
| Audio | AI-generated TTS with per-article audio files |
| Database | MariaDB (Ghost) + SQLite (API layer) |
| Reverse proxy | Nginx with Let's Encrypt SSL |
Agents, swarms, and the pace of change
New research analysis published regularly. If you're building with AI or just trying to make sense of where things are heading, have a read.