Advanced search & filter combinations
When you combine natural-language search with metadata filters (country, date, IPC/CPC, document type, ...) you can pinpoint patents with surgical precision. When prompting an LLM, just list these conditions conversationally.
Filters you can combine
Country
Pick from US · KR · JP · CN · EP. Multiple values allowed.
Example
"Only documents filed at the US and Japan patent offices..."
Doc type
Filter by application status — pending application vs. granted patent.
Example
"Only granted patents (Grant) — exclude pending applications..."
Year / Date
Specify a year/date range using either filing date or publication date.
Example
"Among patents published after 2020..."
IPC / CPC
Narrow to a specific industry or technology area. Prefix-only input (e.g. H01M) automatically covers all sub-codes.
Example
"Among documents with classification code H01L (semiconductors)..."
Practical query combinations
Below are best-practice patterns for what to write in an LLM chat (Claude, etc.) — and how those map to REST API calls.
Send filters as top-level fields
The search API silently ignores fields it does not recognize. Wrapping filters in an object such as filters, or misspelling a name, still returns 200 OK — but with that filter not applied. Send every filter as a top-level field, as in the examples below.
Scenario 1. Tech + specific country + period filter — findip_search_patents
AI prompt (natural language)
"For tech that uses deep learning to remove reflected-wave noise in LiDAR sensors, find the top 5 documents granted/published at the US and KR patent offices from 2022 onward."
{
"query": "deep learning approach to remove reflected wave noise in LiDAR sensors",
"top_k": 5,
"countries": ["US", "KR"],
"publication_date_from": "2022-01-01"
}Scenario 2. Company comparison + IPC narrowing + stats — findip_analyze_trends
AI prompt (natural language)
"Among patents filed by Samsung Electronics and TSMC, find documents in IPC class H01L (semiconductors) related to foundry process-node yield improvement. Then compare their annual filing volume over the last 10 years in a table."
{
"query": "foundry process-node yield improvement",
"applicant_entity_ids": [
"SAMSUNG ELECTRONICS CO LTD",
"TAIWAN SEMICONDUCTOR MFG CO LTD"
],
"ipc_codes": ["H01L"],
"date_range": { "from": "2015-01-01", "to": "2024-12-31" },
"date_type": "filing",
"stats_group_by": ["year", "applicant"],
"min_score": 0.4,
"page_size": 20
}Applicants must be resolved first with findip_find_applicant, then passed to applicant_entity_ids as canonical entity ids — a raw company name will not work. Searching "TSMC", for example, returns only the China and Nanjing subsidiaries; the parent is a separate id, TAIWAN SEMICONDUCTOR MFG CO LTD. When a company has several name variants, pass all of them in one list.
min_score is lowered here because the default of 0.65 is strict: with these filters the default returns 0 results, while 0.4 matches 328 documents. Lower it when results are too few, raise it when unrelated documents creep in.
Scenario 3. Status filter (granted only) — findip_search_patents
AI prompt (natural language)
"For mechanical structures that flatten the hinge crease on foldable smartphones, find only granted patents — exclude pending publications."
{
"query": "structure that flattens the hinge crease on foldable smartphone displays",
"top_k": 5,
"doc_type": "grant"
}