Neuron Press
NeuronPress is an AI-assisted publishing system for technical AI news and analysis.
Visual system
Project Frames
A focused gallery of interface states, art direction, and launch-ready composition details.




Case study
Structural logs & methodology.
The build notes below outline the design intention, engineering priorities, and product decisions behind this specimen.
Start a similar buildEditorial Infrastructure for AI-Native Publishing
NeuronPress is a technical AI publication built around an agentic production workflow. It combines a public React reading experience, Convex-backed content operations, protected editor tools, and a TypeScript publishing pipeline that turns topic discovery, research, visuals, drafting, release, and distribution into one coordinated system.
What It Does
- Finds emerging AI topics from live discovery sources, then falls back to curated evergreen strategies when the signal is thin.
- Selects SEO-aware angles, validates category fit, and carries keywords forward into the writing and publishing stages.
- Generates featured imagery, optional data-visualization blocks, long-form drafts, metadata, Convex posts, and static sitemap updates.
- Stores posts, categories, newsletters, ad slots, agent settings, social campaigns, and social publishing attempts in Convex.
- Promotes new articles through an X-focused Playwright agent while keeping social failures non-fatal to article publication.
- Runs MetaAgent review after publishing, with quality gates around word count, metadata length, image availability, and grounded research.
Technical Architecture
The public site is a React 18 and Vite application styled with Tailwind and shadcn/Radix UI, with root-level article routes, category pages, authentication, and an admin dashboard for posts, ads, and agent configuration. Convex Cloud provides the data model, role checks, post APIs, storage-backed featured images, newsletters, agent settings, and social publishing records. The agent runtime is TypeScript: TrendScout, TopicPicker, Researcher, ImageGen, DataViz, Writer, Publisher, SocialMediaAgent, MetaAgent, and Mechanic share pi-based LLM utilities, structured tools, run logs, deployment scripts, Telegram notifications, and Cloudflare Pages-oriented sitemap automation.
Outcome
NeuronPress shows how an editorial product can be engineered as a disciplined publishing machine: source discovery, content strategy, AI-assisted production, storage, admin review, SEO hygiene, deployment, observability, and social distribution are treated as connected infrastructure rather than disconnected blog tasks.