Back to demo

How It Works

Find recent papers, extract relevant passages, and answer a research question with citations.

Why Exa

Research discovery needs semantic relevance, recency, and enough context to distinguish a title match from a substantive contribution. Exa searches publication sources and returns focused excerpts for comparison and citation.

The Pipeline

Overview

The demo performs a live Exa search, optionally augments arXiv results with citation counts from OpenAlex, then streams a short synthesis based on the five leading papers.

Steps

1

Search publication sources

POST /search uses type auto, category publications, and a result count capped at 15. A selected week, month, three-month, six-month, or year window becomes startPublishedDate; the all-time option omits it.

/search

2

Clean paper metadata and enrich citations

The app asks for up to 4,000 characters of relevant highlights per result, cleans titles and abstracts, derives missing arXiv dates, and tries a five-second OpenAlex lookup for citation counts. OpenAlex failure never blocks results.

3

Summarize the five leading papers with citations

The top five results go to a synthesis model, which defaults to gemini-3.7-flash. The prompt asks for a 250-word summary and citation markers, which are mapped back to the result cards.

4

Look up a researcher on demand

Selecting a researcher triggers a separate Exa Search for that name plus the word researcher. This call sets category people and numResults 3, and requests the full page text. It leaves type unset, so Exa uses its default.

/search

Endnote

OpenAlex is an optional enrichment layer and is not part of the Exa request. Citation count is therefore nullable, while every displayed paper still comes from the Exa publications search.