Personal project2026
byte
Tech in bite-size — a TikTok-style vertical feed of new dev tools, aggregated from 6 sources and ranked so the fresh + well-liked stuff floats to the top.
Context
byte is a full-screen, snap-scrolling feed of the tools, libraries, and projects the dev world is liking right now — one card per item, infinite scroll, like TikTok for tech.
It aggregates six sources (Hacker News, GitHub, Dev.to, Lobsters, Product Hunt, and more) into one normalized feed, then ranks and dedupes so the freshest well-liked items surface first.
What I built
- Six source adapters (Algolia HN, GitHub Search, Dev.to, Lobsters, Product Hunt) normalized behind one `FeedItem` model — add a source by dropping in one file.
- Trending rank via the classic HN gravity shape: popularity damped by `log`, decayed by age; cross-source dedupe.
- Card enrichment fetches each link's `og:description` and `og:image` so every item explains what it is and shows a visual.
- Keyword classifier categorizes items (AI, Dev Tools, Web, Data & Infra, …) with no LLM — instant.
- First-run interest picker (saved to `localStorage`) drives a personalized For You view; server-side pagination gives effectively infinite scroll.
Deep dive
Fast without an LLM or a database
The whole feed is server `fetch` with ISR (30-min revalidate) streamed via `loading.tsx`, so prod serves the cached feed near-instantly and the animated splash holds only until real content paints. Ranking, dedupe, and categorization are plain functions — no vector store, no model call — which keeps each request cheap and the classifier instant.