Patent Embedding Radar: Spotting Competitor Encroachment

Filing Stream

New filings 1.2 / s

Encroachment Detection

Similarity threshold 0.90

Filers

Competitor A0

Competitor B0

Competitor C0

Your portfolio0

Live Readouts

Total filings

Encroachments

Closest threat

–

Whitespace gap

–

How it works

Every filing is embedded as a unit vector on a similarity sphere (a 3D stand-in for the hundreds-of-dimensions text embeddings a real NLP pipeline would produce from patent claims). Vectors from the same technology cluster point in similar directions:

sim(u, v) = (u · v) / (|u| |v|) = cosθ   (u, v unit vectors)

An encroachment is flagged whenever a competitor's filing direction has cosine similarity ≥ threshold to your nearest filing — geometrically, it lands inside the cone around one of your patents:

encroached(c) = max_i sim(c, you_i)  ≥  threshold

The whitespace gap samples ~180 probe directions spread evenly over the sphere (a Fibonacci lattice) and finds the probe that is farthest — in angular terms — from every existing filing:

gap = max_p ( min_i  arccos(sim(p, filing_i)) ) × 2

That probe marks the largest unclaimed region of the technology space — a candidate whitespace opportunity, shown as the gold marker. Real competitive-intelligence platforms run this exact nearest-neighbour logic over genuine text/patent embeddings (often 300–1500 dimensions) instead of 3 — the sphere here makes the same math visible.

Filing stream — new patents arrive continuously from you and three competitors, each pulled toward one of six technology clusters with directional noise.
Threshold slider — tightens or loosens what counts as "too close" to your IP.
Filer toggles — hide a competitor to see how much of the detected threat comes from them specifically.

Your portfolio Encroaching filing Whitespace opportunity

⚠ Couldn't load the 3D engineThree.js failed to load from the CDN. Check your connection and reload.