Engineerious
A text-first AI engineering desk that combines reviewed intelligence, technical writing, and a human-controlled research pipeline.
Problem
Important AI developments arrive through lab announcements, research papers, repositories, engineering blogs, and media. A fast feed can report what changed without helping an engineer decide what deserves attention.
The product therefore has two jobs: preserve useful collection and ranking infrastructure, then add the context, technical significance, and human judgment that a generic aggregator cannot provide.
Why it exists
Tharun is building Engineerious as his personal AI engineering desk: one place for practical writing, reviewed AI intelligence, project notes, and the questions he is working through.
The editorial goal is straightforward. A useful entry should explain what happened, why it matters, what changed technically, and what an engineer should learn from it. Volume is not the measure of success.
Architecture
The sequence keeps collection useful without letting automation cross the editorial boundary.
- Collect. Source adapters gather candidate news, model, open-source, and research material from public feeds and APIs.
- Normalize and rank. Candidates share one item model, deduplicate by canonical URL hash, and use a Hacker News-style score with source authority and freshness.
- Research and draft. The daily workflow turns gathered material into source-linked findings and an unpublished database-backed draft.
- Review. Grounding and voice checks surface problems for the reviewer. AI-generated work remains a pending draft until a named human approves it.
- Publish. Verified MDX writing and manually approved database-native work become public; collection never implies publication.
Engineering decisions
Separate collection from publication
Ingestion can continue behind the scenes while unfinished or unreviewed feed surfaces remain closed to the public.
Use two deliberate content sources
Human-authored writing stays in typed MDX. Runtime daily drafts use Postgres because a deployed server process cannot write durable files back into the repository.
Make provenance part of the data
Published writing tracks origin, source status, testing status, authenticity, reviewer, and review time instead of treating trust as presentation copy.
Fail closed at editorial boundaries
Generated source URLs are checked against gathered material, protected routes require credentials, and AI-generated drafts cannot mark themselves verified.
Current work
- Strengthening the personal publishing surface around Tharun's writing, projects, and engineering point of view.
- Turning the daily research workflow into a concise, reviewable intelligence experience instead of another long automated post.
- Opening curated topic and handbook surfaces only when they contain reviewed material worth publishing.