Skip to content
ibruno
All applications

isurco

An engine designed so that you write a technical problem in plain language and receive solutions drawn from expired patents, forgotten science and theses, each with the verbatim quote that backs it and a confidence traffic light. If there is no proof, there is no card.

Eight lenses and a quote that is not there

Example problem · a lightweight insulation for facades that does not burn

  • articles
  • free patents
  • ahead of their time
  • forgotten science
  • theses
  • nature
  • analogies
  • gaps

If the quote does not appear verbatim in the original text, the claim is withdrawn.Example · sample data

On screen

Screenshots of the interface prototype, with sample data

isurco1 / 3
Simulated search: the lenses light up one by one as the first findings with a verified verbatim quote appear; the prototype shows six of the eight lenses and both filters (interface in Spanish). Sample data

What it is

Looking for a technical solution usually ends in a list of documents you must read, compare and then trust. And much of the useful knowledge sits where nobody looks: patents that have already expired, old untranslated science, doctoral theses. isurco is designed to start from a single text box: you describe the problem and the system is meant to return solutions, not documents. Each card links to a concrete passage of a source; if the quote does not appear verbatim in the original text, the claim is withdrawn. And the system also says the uncomfortable part: whether something was already tried and failed, whether it has been reproduced, or that a patent may still be in force.

Who it's for

Someone with an engineering or product problem who wants precedents with proof. It is a personal project, on hold today: this page describes how it is built; it is not offered as a service.

How it works

  1. One box and, at most, one question

    The user writes the problem; there are no modules, sources or filters to choose. If it is ambiguous, the system asks a single question with options, never more than one, and the search can go ahead anyway.

  2. Lenses that search in parallel

    Eight lenses share one interface: articles, free patents, ideas ahead of their time, forgotten science, theses, nature-inspired solutions, analogies between fields and gaps in the literature. Two filters — prior failures and reproducibility — complete the picture. Cards are streamed live, as server events, as they are found.

  3. Nothing without a verbatim quote

    A card is stored only with at least two verified pieces of evidence, from independent passages, and every quote must be contained in the original text. Violations are blocked by the database itself, not only by the application.

  4. A traffic light that measures the proof

    Green, amber, red or grey according to the strength of the evidence, the confidence and any open failures; not according to whether the solution suits you, which “Why it fits” explains. For patents, freedom to operate is shown only as guidance and the interface never says “free to use”.

  5. Honest when there is no proof

    If no candidate gathers verbatim, independent quotes, the screen says so — “Not enough proof” — and reports how many candidates it reviewed and discarded, instead of filling the gap with something plausible.

Scope

Real status and scope:

  • The project has been on hold, by its author's decision, since 7 November 2026. It is not deployed or offered in any environment.
  • There is no full corpus: bulk loading is out of scope and only a local development sample has been ingested. Sources start disabled in the database and are enabled only after each one's terms of use have been reviewed.
  • In tests with real data, one lens (forgotten science) worked end to end, but the cards came from a single document: corroboration across several documents is still unproven. The free-patents and theses lenses have not yet produced cards with real data.
  • Spanish is the product language, and it is designed for individuals, not for teams or organisations (out of scope). The language model is local; problems do not leave the machine unless configured otherwise.

On hold since 7 November 2026, at its author's request. Built with adversarial review and mutation testing; still without a full corpus or delivery.