Welcome to IPZilla.
Patent-grade IP intelligence for the AI-drug-design era.
IPZilla turns global patent landscapes into bulletproof, click-to-file provisional and full patent applications. It pairs an AI patent landscape engine with a Claim Interpretation Lattice, then applies a legal cartography that neutralises 69 documented failure modes drawn from US §112 (Amgen v. Sanofi), EPO G 2/21 plausibility, the USPTO Inventorship Guidance of November 2025, the UPC Court of Appeal UPC_CoA_528/2024, In re Cellect, and the rest of the post-2023 case law that quietly rewrote how pharma patents survive litigation.
The product ships as three modules. Each one solves a different problem, and each one feeds the next.
IP Radar
Patent landscape engineReads a one-paragraph brief, plans a multi-strategy search across the 130M-patent Google Patents Public Datasets, fans out per-patent claim-reading via Gemini Pro, and returns a reranked landscape with element-wise FTO verdicts. Already built — this is the surface most of the API exposes today.
Claim Interpretation Lattice (CIL)
Trade-secret claim moatEnumerates 10⁵–10⁸ interpretations of every claim across six axes (literal, doctrine of equivalents, Markush expansion, jurisdiction, era of jurisprudence, prosecution history), cross-references each one against the global prior art, scores its vulnerability, and proposes the wording that maximises cross-jurisdictional robustness. Spec'd in DOC/MODULE_CIL_SPEC.md; MVP in flight.
Doctrinal Hardening
The legal cartography appliedWraps every output of CIL in the cascade-4-couches (functional genus → structural genus → ranges → picture claims), inserts plausibility statements for G 2/21, doubles each functional claim with a structural anchor for Amgen, watermarks numerical examples, and cross-files preemptive terminal disclaimers for In re Cellect.
The workflow is deliberately small. You drop a brief — "amorphous solid dispersion of indomethacin in PVPVA" is enough — and IPZilla runs the loop:
- Brief. Free text in. We parse it into a structured
DrugBriefobject. - Landscape. Multi-strategy SearchPlan, BigQuery against 130M patents, family dedup, LLM rerank.
- Claim. Per-patent claim fetch (USPTO PEDS / EPO OPS), claim-reading via Gemini Pro, ClaimExtraction cached globally.
- Screen. Element-wise FTO verdicts per patent. Holistic verdict on top.
- File. CIL drafts the cascade, the cartography hardens it, the export pipeline ships a ZIP and an INPI-provisional-ready PDF/A.
End-to-end, the first four steps complete in roughly eight minutes for a typical brief. That's the same horizon a cabinet IP analyst takes to read the first three patents — except IPZilla has read every patent that intersects the scope, with a ClaimExtraction cached for the next customer who comes along.
What you should expect from these docs: precise, sourced, and deliberately un-marketed. Every page below is generated from the same 17-section playbook (DOC/00_PLAYBOOK_MASTER_SOTA_12M.md) that drafts your provisionals. We cite case names with their F.4th reporters, we name the EPO Boards of Appeal decisions by their G/T/J number, we link the USPTO MPEP sections by paragraph. If a paragraph here is wrong, it is wrong in the patents we draft. We treat both as load-bearing.
What you should not expect: a substitute for a CPI on the first deposit at high commercial stakes. IPZilla is calibrated to do 90–95% of what a cabinet IP team does, automatically, and to do it well. We recommend a registered mandataire for the final read-through on PROV1 and on the consolidated PCT — not because we don't trust the output, but because the residual risk envelope of self-drafting at ten-million-euro stakes is not zero. The pages on legal cartography and the provisional patent guide are explicit about which residual risks are not coverable algorithmically as of May 2026.
If you're new, the next stop is the Quickstart — six steps from signup to a downloadable INPI-ready PDF. If you're an engineer, jump to the API reference. If you want to see the full doctrinal map first, open the legal cartography page.