Mark by AirtopMeet Mark. Vibe automation for marketers.
Promoted
AI Research Lab
Read 500+ AI papers without reading 500 papers.
4 followers
Read 500+ AI papers without reading 500 papers.
4 followers
AI Research Lab is a multi-agent observatory over arxiv. 5 specialized LLM agents read 502 papers and surface what matters across them — 39 contradictions, 48 consensus findings, 19 open debates, frontiers. Weekly fresh ingest, every step documented.
Hey PH — Maker here. Software engineer by background. I built AI Research Lab over a few weekends using an AI coding agent as my development environment — returning to the keyboard with an AI pair.
What it actually does: ingests arxiv papers, runs 5 specialized LLM agents (summarizer, contradiction-finder, frontier-detector, trend-mapper, benchmark-extractor), and renders the output into a 5-tab UI per research topic. The tab I care about most is Insights — where the contradiction-finder and trend-mapper outputs land, so you can see "these 4 papers disagree about X" instead of reading 4 abstracts in sequence.
The numbers: 502 papers across 9 topics, 7,393 citations mapped, 837 flagged as influential, 39 contradictions / 48 consensus findings / 19 open debates / 8 warnings already extracted.
Stack: Next.js 15, Postgres + pgvector, Gemini text-embedding-001, an instruct-tuned LLM, Cloud Run. Weekly cron. RSS feed per topic. Open source (MIT).
What I'd love from PH: try a topic, click into Insights, tell me if a contradiction is wrong. If you're a researcher: which agents are missing? If you're an operator or ex-engineer using an AI coding agent to ship things — DM me.
Try it: https://www.airesearchlab.space/
Code: https://www.github.com/abhid1234/AI-research-lab
Blog: https://abhid.substack.com/p/wha...
I'll be here all day. — Abhi