PaperSift

What PaperSift is

PaperSift is a search engine over the published proceedings of the major AI, machine learning, computer vision and medical imaging venues. Every paper's title, authors and abstract is indexed in one place, so a single query reaches across all twelve venues at once instead of twelve separate conference websites.

It runs entirely in your browser. There is no account, no login, no query logging and no backend server — the index is downloaded once and searched locally, which is why results appear as you type. It is free and open source under the MIT license.

How it helps researchers

• One query, twelve venues. Start a literature review without opening NeurIPS, CVPR, MICCAI and nine other proceedings sites in separate tabs — each with its own search box, its own quirks, and no way to compare across them.
• Follow a person, not just a topic. The author: prefix pulls up everything a researcher or lab has published across the whole archive — useful for tracking a group's trajectory, preparing to review, or collecting your own record.
• See where a field actually publishes. Filter by venue and year, and the filter chips reorder by how many matches each venue holds — so a query answers "who works on this, and where do they send it?" as well as "what exists?".
• Triage without opening PDFs. Abstracts expand inline with matched terms highlighted, and mathematical notation renders properly via KaTeX, so you can judge relevance from the results list.
• No waiting. Because the whole index sits in the browser, filtering 146,000+ papers takes milliseconds and works the same on a slow connection once loaded.

What is indexed

Twelve conferences and journals, covering every year listed below. Counts come straight from the live index, so this table updates whenever the archive is re-synced.

How to search

A phrase — just type it

knowledge distillation finds papers whose title or abstract contains those words side by side, in that order.

Both terms — and

classification and calibration finds papers that mention both, anywhere in the title or abstract.

Either term — or

classification or calibration finds papers that mention either one. When you mix them, and groups first: mri or ct and segmentation means mri, or ct together with segmentation.

By author — author:

author: abhishek sambyal lists that person's papers. A space keeps the parts of one name together, and a partial name such as author: sambyal works too.

Several authors — or / and

author: aleksei tiulpin or abhishek sambyal lists papers by either person; author: aleksei tiulpin and abhishek sambyal lists only the papers they wrote together.

Authors and topics together

Keep going with and / or: author: abhishek and bathula and calibration finds calibration papers the two wrote together. After author:, any word that is someone's name is read as a name and everything else as a topic; the results line shows which is which. A few topic words are also surnames (gan, brain) — put those before author: to keep them topics: gan and author: hinton.

Narrowing by venue and year

The Venue and Years chips filter results. Leaving a query empty and selecting only a venue and year browses that whole proceedings.

Where the data comes from

Every record is collected from the venue's own published proceedings, or from DBLP and OpenAlex where a publisher does not expose abstracts: papers.nips.cc (NeurIPS), proceedings.mlr.press (ICML, MIDL), openaccess.thecvf.com (CVPR, ICCV), ecva.net (ECCV), ojs.aaai.org (AAAI), ijcai.org (IJCAI), papers.miccai.org (MICCAI), iclr.cc (ICLR), and DBLP + OpenAlex (ISBI, TMI, and older MICCAI/MIDL/ICLR years).

Coverage is honest about its gaps rather than silent about them:

• AAAI is best-effort — its OAI-PMH feed paginates unreliably under load, so some years may be incomplete.
• MICCAI years sourced from DBLP have no abstracts; Springer does not expose them and blocks scraping. 2024 onward has full abstracts.
• ICLR abstract coverage is roughly 81% — some detail pages carry no abstract to collect.
• MIDL has no 2018 entries (DBLP has none) and nothing for 2025 onward yet.

Titles, authors and abstracts are shown for discovery only; the papers themselves stay with their publishers, and each result links out to the original source.

Who built this

PaperSift is built and maintained by Abhishek Singh Sambyal, a researcher working in machine learning and medical image analysis. It started as a way out of a routine that will be familiar to anyone doing a literature review: opening one conference site after another, running the same query in a dozen different search boxes, and still not knowing whether anything had been missed.

More at abhisheksambyal.com — missing venues, wrong records and feature ideas are all welcome.