Featured Project

AI-Maintained GPU Cloud Pricing Directory

A GPU cloud pricing directory covering 31 providers and 1,000+ LLMs, kept current by an autonomous AI agent pipeline that collects, validates and publishes prices daily.

View Live Project
nodepedia homepage comparing GPU cloud prices across providers, with a live price ticker.

Technologies Used

PythonHugoTailwind CSSPlaywrightPagefindGeminiDeepSeekRaspberry PiVercelPlausible Analytics

About This Project

The Challenge

GPU cloud prices change constantly across dozens of providers, each publishing them in a different format. A comparison site maintained by hand goes stale within weeks. The goal was a directory that keeps itself current without daily manual work.

The Solution

nodepedia is a static Hugo site with no database: all data lives in version-controlled YAML and Markdown. An autonomous Python agent pipeline updates it daily. It collects prices, validates them, writes the content and commits to git, and each commit deploys automatically to Vercel. The site compares 31 providers, 86 GPU models and over 1,000 LLMs, and adds tools such as a VRAM calculator, a cost calculator and a model matcher.

Technical Highlights

1. Multi-Agent Architecture

An orchestrator assigns work through a file-based task queue to two specialized agents. A fast, low-cost model (Gemini Flash Lite) handles extraction. A stronger model (DeepSeek) writes content. Every model call and every run has a hard cost cap.

2. Scraping with Guardrails

  • LLM-driven Playwright browsing for provider pricing pages, plus direct API integrations for GPU marketplaces
  • Prompt-injection guards, price-bound checks and anomaly detection, applied before any number reaches the site
  • Price history, plus rebrand detection with automatic URL redirects

3. Self-Hosted on a Raspberry Pi

The pipeline runs on a Raspberry Pi 5, started by a systemd timer, with secrets injected from 1Password at runtime. A health gate and a dead-man's-switch heartbeat send Slack and Discord alerts if a run fails or never starts.

4. Search and SEO

Structured data (schema.org JSON-LD) for products, FAQs and breadcrumbs, automatic submission of new pages to the Google Indexing API batched within the daily quota, and fast client-side search with Pagefind.

5. Testing

The agent codebase has about 2,700 pytest tests and is checked with ruff and strict mypy.

Project Gallery

NVIDIA L40S page with specs and a table of rental prices by provider.

GPU pages combine specs with live prices and price trends from every provider.

GPU cloud pricing matrix comparing GPU models across providers.

The full price matrix, with daily price movements shown by arrows.

RunPod provider page with scores and a GPU pricing table.

Provider pages with scores, a verification date and the full GPU price list.

Interested in working together?

Let's discuss how I can help bring your project to life.