Natural-language search
Findip is a semantic search engine. The more you frame your query as a sentence describing intent and the problem to solve — rather than a few keywords — the better the results.
This page covers the basics that apply to every workflow. For role-specific deep dives (prior-art, FTO, invalidity, landscape), see the cards near the bottom.
Write a concrete sentence, not a keyword list
Semantic search doesn't look for documents that share words; it finds the documents whose meaning is closest to your question. Even when an exact technical term is missing, the engine still surfaces patents that describe the same problem-solving principle or use a synonym.
Not recommended — keyword soup
"all-solid-state battery lithium dendrite suppression solid electrolyte"
- Lacks context, so it falls back to plain word matching
- Misses patents that say "dendritic crystal" instead of "dendrite"
Recommended — sentence with intent
"Solid-electrolyte technology that suppresses lithium dendrite growth in all-solid-state batteries to improve stability"
- Goal (stability) and challenge (suppress growth) are explicit
- The vector engine cleanly clusters patents that share the same intent
Powerful applicant-aware search
When you prompt an LLM to find a company's patents, you don't need to know the exact English / Korean / Japanese name. Findip links tens of thousands of applicant-name variants to a single representative Entity ID under the hood.
Example
Prompt
"Find Apple's recent patents on augmented-reality (AR) HMDs."
Given that prompt, the LLM resolves the applicant first, groups name variants such as Apple Inc., 애플 인코포레이티드, and アップル インコーポレイテッド under one canonical id, then searches US · KR · JP · CN · EP in a single pass.
When a name maps to several legal entities, though, the choice is not automatic. A parent company, its local subsidiaries, and affiliates (Electronics vs Display vs SDI, say) are separate ids — so if you have a specific scope in mind, state it in the prompt.
Search vs Trends
Switching the way you prompt the LLM (which endpoint gets called) based on what you actually want from the answer makes a huge difference in efficiency.
1. When you need deep dives into individual patents
Use this when you want to find the most similar patents in a technology area and read their content.
Example prompt
"Summarize the abstract and claims of the 5 most similar core patents that improve the interfacial resistance of solid electrolytes."
2. When you want market-wide trends
Use this when you need to skim thousands of documents fast for stats like annual filing volume or top-applicant rankings.
Example prompt
"Give me a table of yearly filing volume for autonomous-driving LiDAR sensors over the past 5 years and the share of the top 5 companies."
Guides by work type
Different goals call for different prompting styles, filters, and tools.