AI Automation · In-House Product

FoundersDev

The engine behind this site's blog: on a schedule, it picks a founder topic or reads a new piece from the writers founders trust most, drafts a practical article in my voice, puts it through an editorial review, and publishes it — with no manual steps.

Category
Autonomous Research, Writing & Publishing Pipeline
Status
Live
Core stack
Gemini 3.8 Flash, TypeScript, rss-parser + Cheerio, GitHub Actions
Live at
samuelstanley.com/blog
samuelstanley.com/blog
FoundersDev — The archive
628
Articles published

Playbooks, essays and earlier dispatches, all generated and committed by the pipeline

48
Founder topics

Across six pillars of running a company

0
Manual steps

From research to a live page, on a schedule

01The challenge

Writing genuinely useful material for founders every week takes research and editing time that client work can't spare, and generic AI content reads like filler that costs more trust than it earns.

02The approach

A scheduled pipeline that treats writing like a product. It works through a curriculum of founder problems, reads recent essays from Paul Graham, First Round Review, Lenny Rachitsky, YC and African operators, drafts with Gemini against a strict voice and evidence guide grounded in real project history, then runs a second editorial pass and a quality gate before committing Markdown that the site publishes automatically.

03What I built

The pieces that make it work.

01Editorial Planning

Founder Curriculum

Six pillars (strategy, growth, operations, product and tech, team, finance) and dozens of topics, rotated so the blog covers the whole job of running a company rather than the news cycle.

02Research

Reads the Best Founder Writers

Pulls recent pieces from eleven trusted sources, skips announcements and podcast promos, and writes a response that credits the original as further reading.

03Two-Pass Generation

Draft, Then Edit

A second model pass acts as editor: it removes invented evidence and client claims, turns listicles into prose and enforces the house voice.

04Reliability

Quality Gate & Fallbacks

Drafts that are too short, badly structured or full of filler are rejected; rate limits and unavailable models fall through to the next model automatically.

05Publishing

Git as the CMS

Each post is a typed Markdown file committed to the repository, so every article has history, review and rollback for free.

06Operations

On-Demand Runs

An admin can point the engine at any article or video URL, or run a dry run that previews a post without publishing it.

04Inside the product

Real screens, in production.

samuelstanley.com/blog
The archive
The archiveSearch, topic filters and every published article.
samuelstanley.com/blog
A published essay
A published essayReading layout with sharing, tags and further reading.
Mobile archive
Mobile archiveThe blog on a phone.
05Architecture

How it's put together.

  1. Scheduling

    Scheduled CI Workflows

    Two cron workflows: practical playbooks twice a week and a weekly essay. Each run commits its post and its history file so topics never repeat.

    • GitHub Actions
    • Node.js
    • tsx
  2. Research

    Feeds and Page Extraction

    RSS feeds and page scraping turn recent essays into clean text, with dedupe history so the same source is never answered twice.

    • rss-parser
    • Cheerio
    • TypeScript
  3. Writing

    Two-Pass Generation

    A voice guide built from real portfolio data drafts the post; an editor pass enforces evidence and style rules; a parser builds the frontmatter in code so it is always valid.

    • Gemini 3.8 Flash
    • Google Generative AI SDK
  4. Publishing

    Markdown to Static Pages

    Committed posts trigger a deploy; Next.js renders the archive, search, topic filters and article pages, and refreshes hourly.

    • Next.js
    • gray-matter
    • marked
06Stack decisions

Every tool earns its place.

Gemini 3.8 Flash
AI
Drafting and editing
Fast, low-cost long-form generation with enough context for a full source essay.
TypeScript
Language
Pipeline logic
Typed parsing and validation keep a bad draft from ever reaching the site.
rss-parser + Cheerio
Research
Reading sources
Lightweight feed and HTML parsing without running a browser.
GitHub Actions
Scheduling
Cron and commits
Runs on a timer with no server to keep alive.
Next.js
Framework
Rendering the blog
Static article pages with hourly revalidation as new posts land.
07Engineering notes

Decisions worth writing down.

01

Voice from facts, not adjectives

The writing guide is built from real project data, so stories about my own work stay true and specific instead of invented.

02

An editor pass beats a longer prompt

First drafts bend rules — invented benchmarks, listicles, name-dropping. A second pass with one job, enforcing the rules, fixed more than any amount of prompt tuning.

03

Frontmatter in code, not in the model

The model returns plain fields; code writes the YAML, so a stray quote can never break a page.

Building something like FoundersDev?

Next case study

Skyways Shipping→

Logistics · Web + Mobile Platform