SYED_ABBAS
← All work
CASE STUDY · [04]
YEAR · 2024
ROLE · Systems / GPU

Local AI Infrastructure

Home rig for LLM orchestration, distributed scraping, and video generation. Everything runs locally.

Outcomes
0Scraper devices
0Orchestration
MEDIA SLOT
rack / GPU shot, or dashboard screenshot
BRIEF

Cloud LLM bills add up fast when running long research jobs. I wanted a rig I could point at a question and let run overnight without watching a token counter.

[01]

What the rig runs

A central controller distributes jobs across local GPUs. A scraper fleet of Raspberry Pis, a Jetson, and a Windows storage node collects and dedupes data on a schedule. When a local model can't handle a request, the fallback ladder goes to Groq, then Gemini, then OpenRouter, chosen by cost and rate-limit headroom.

MEDIA SLOT
pipeline dashboard
MEDIA SLOT
device topology diagram
[02]

The scraper fleet

Reddit collection rotates hourly between JSON endpoints, PullPush, and ArcticShift so no single method hits its rate limit. App Store, Play Store, Twitter, and GitHub sources run from a Mac cron at staggered hours.

STACK
PythonvLLMNode.jsSQLiteRaspberry PiGroq/Gemini fallback