Skip to content
Prajwal Bhatia
All work

Virtual Internships · 2025

Thumbnail Pipeline

Most portfolio links in the candidate browser had no uploaded image, so rows showed the same placeholders. I built a backend service that derives previews from those links without changing the frontend or slowing its API response.

Before and after

The same row with generated previews of both linked projectsBrowse row with two generic placeholders instead of project previews
Candidate name, institution and previous employer redacted

How the pipeline works

The API first checks Redis and a URL-hash database record. On a miss it returns a placeholder immediately and queues a BullMQ job. The worker reads og:imagewhen available, then falls back to a Puppeteer screenshot. It stores the result for the next request. The existing frontend continues to render the same image_urlfield.

A request returns a placeholder while a worker validates the URL, extracts an Open Graph image or takes a screenshot, then stores the preview for later requests.

Safety and result

Because the links are supplied by users, the worker validates IP ranges and DNS results before fetching. Failed jobs leave the original placeholder rather than failing the browse request. In production the pipeline processed 95% of links successfully; the Open Graph path handled about 55% of URLs in roughly one second, at about one eighth of the Puppeteer fallback's latency.

Interested in the rest of the work?

Browse projects