Problem

Academic and industrial researchers spend substantial time finding, reading, validating, and synthesizing a scattered scientific literature. Epsilon positioned the pain as a workflow problem: its site promised to cut roughly twenty hours of research down to twenty minutes, while its launch description argued that general-purpose LLMs were not well suited to researchers who need evidence and references. The central use case was accelerating literature reviews and evidence discovery for grants, proposals, meta-analyses, prior-art searches, and new research questions.

The most compelling “killer” workflow was asking a focused scientific question and quickly getting a defensible answer grounded in papers rather than an uncited model response. Researchers could also use the system to identify seminal papers, test claims against multiple sources, and learn unfamiliar subfields without manually screening the entire literature.

Product / Service

Epsilon was a web-based AI search and synthesis tool for scientific research. It searched a Semantic Scholar dataset containing more than 200 million academic papers, selected the top 100 papers related to a query, and passed the query and papers to GPT-4 to generate a summary with inline citations. Users could search publications and patents, open papers, save publicly available PDFs, and organize them in personal libraries.

The product extended beyond question answering: it could scan search-result papers for evidence, extract information relevant to a claim, compare findings across documents, generate literature reviews, and summarize uploaded papers or entire libraries. The benefit was a faster research workflow with traceable source material, making AI output more useful for citation-heavy work and reducing repetitive screening and synthesis.

Market

Epsilon competed in the AI-enabled academic and scientific research market, particularly literature search, paper discovery, evidence extraction, and research synthesis. Tracxn identified Flywheel, SciSpace, and JSTOR among its leading competitors, while the broader category also includes newer AI source-finding and document-reading products such as Sourcely and Rorobot. Epsilon’s differentiation was the combination of conversational answers, large-scale academic search, and inline references to the underlying literature.

Historically, the company was a Y Combinator Winter 2023 startup founded by Eshan Agarwal and Viraj Rai, with the company site claiming more than 30,000 researchers and listing use cases across evidence search, citations, grants, literature reviews, and meta-analyses. Tracxn reports a $500,000 seed round backed by Y Combinator and five employees as of April 30, 2026. However, the official site announced that Epsilon would shut down on April 30, 2026, so as of August 1, 2026 it should be treated as a discontinued product rather than an operating competitor. The available evidence does not reliably establish revenue or support a definitive pre-revenue classification; the strongest confirmed signals are user traction and seed funding.