Patent Landscape - prior-art intelligence by full-text mining

Find out what the world has already decided is impossible - and whether that decision still holds.

What it does

Give it a product category. It mines the full text of patent and scientific literature - not abstracts - for concepts that were described as infeasible, impossible or prohibitively slow when they were first disclosed. It then reports which of those barriers have since fallen, and ships the verbatim sentence that proves it.

Abstracts say what a paper claims. Body text says what actually stopped the work. That difference is the whole product.

Example

Input

{
  "queries": ["single cell sequencing"],
  "oldest_year": 2010,
  "papers_per_query": 25,
  "top_concepts": 5
}

Output (abridged, real shape)

{
  "query": "single cell sequencing",
  "corpus": "biomed",
  "papers_read": 25,
  "landscape": [
    {
      "concept": "called infeasible",
      "claim": "...remains computationally prohibitive for routine clinical use.",
      "source": "PMC3520994",
      "url": "https://europepmc.org/article/PMC/PMC3520994",
      "year": 2012,
      "why_it_fell": "Sequencing costs fell roughly six orders of magnitude after this paper. The barrier did not soften - it evaporated."
    }
  ]
}

Every hit carries the quote. A landscape without quotes is an opinion.

Run it

In the Apify console - the example input is pre-filled, so press Run.

From your code

curl -X POST "https://api.apify.com/v2/acts/{actorId}/runs?token=YOUR_TOKEN" \
  -H "Content-Type: application/json" \
  -d '{"queries":["single cell sequencing"],"papers_per_query":25}'

Then read the dataset:

$curl "https://api.apify.com/v2/datasets/{datasetId}/items?token=YOUR_TOKEN"

As an MCP server - the actor is MCP-ready, so an agent can call it directly.

Why the output is trustworthy

Two rules are enforced in the pipeline, both learned the hard way:

1. Reference lists are stripped before matching. Otherwise the barrier phrase matches another author's abstract in the bibliography, and the precision collapses.
2. An empty result is never silent. If nothing matches, it says so and reports what it read - so you can tell "no barriers found" apart from "it did not look".

Limits

• Patent results are low confidence and are not patent law. Treat them as pointers, not conclusions.
• There is no SLA. This is a research tool.
• It needs the source corpus to be reachable when it runs.

License

MIT