How LitHypo works.

LitHypo retrieves papers from Europe PMC, which indexes peer-reviewed journals, PubMed Central, preprints, and patents. We chose Europe PMC because it provides structured metadata (titles, authors, DOIs, PMIDs, journal info, publication dates) through a public API and includes preprints alongside peer-reviewed work — important for fast-moving fields like synthetic biology.

We do not use Semantic Scholar, Scite, Google Scholar, or our own scraped index. All citations in LitHypo output are retrievable through Europe PMC URLs.

For each query, LitHypo:

1. Constructs a structured Europe PMC search query from your topic, expanding it with relevant synonyms and concept groupings where appropriate.
2. Retrieves up to 15 of the most relevant recent papers matching your query.
3. Passes the retrieved paper metadata (titles, abstracts, authors, journals, years) to Claude Opus 4.7 — the strongest currently available reasoning model — with a structured prompt requesting five falsifiable hypotheses anchored in those specific papers.
4. Returns the hypotheses with their supporting paper citations, experimental outlines, confidence ratings, and identified risks.

The engine is grounded in literature you can verify. It is not pulling facts from training data. It is synthesizing across real, retrievable papers.

LitHypo is not a literature search tool. If you want to find all papers on a topic, use Europe PMC directly or a dedicated search tool like Consensus.

LitHypo is not a fact-checking tool. It synthesizes hypotheses from retrieved papers but does not verify the claims within those papers.

LitHypo is not a replacement for reading primary literature. The hypotheses it generates are starting points for research, not conclusions.

LitHypo does not have access to your unpublished work unless you explicitly upload it (in which case it stays private to your account).

Sparse-literature queries: Very narrow or technically specific queries (e.g., "phospho-sensor for engineered CDKA-Cyclin A in mouse embryonic stem cell assembloids") may return zero matching papers. When this happens, the engine refuses to generate hypotheses and tells you to broaden your search.

Hyphenated technical terms: Hyphenated phrases like "phospho-sensor" or "phase-separation" can confuse literature search systems that treat them as single tokens. The engine attempts to expand these with synonym groupings, but very narrow combinations may still return zero results.

Fast-moving subfields with sparse literature: If a research area is so new that fewer than 5–6 papers exist on it, LitHypo will generate hypotheses from the available papers but the synthesis will be limited. We surface the actual paper count to the user.

Confidence calibration: Confidence ratings (HIGH, MEDIUM, LOW) are the model's estimates based on the strength of the literature evidence. They are not statistical predictions and have not been validated against actual experimental outcomes.

Over time, LitHypo will track which hypotheses generated by the engine were tested by users and how those experiments resolved. This longitudinal data — connecting hypothesis generation to real experimental outcomes — does not exist in any other AI research tool. As we accumulate it, the engine's confidence calibration will become empirically grounded.

This is a long-term project. We are at the beginning of it.

Your queries, your uploaded files, your hypothesis history, and your outcome notes are private to your account. We never share your data with third parties. We never use your uploads or queries to train AI models. Other users never see your work. You can delete anything at any time.

We retain anonymized aggregate metadata (queries per day, average response time, confidence distribution) for product improvement, but not the content of any specific query or output.

LitHypo has specialized verticals built for specific research disciplines. For psychology and psychiatry research → https://psychhypo.com