Patent Feasibility

Your best inventions are
already in your systems.

Engineers rarely stop to file invention disclosures — the innovation lives in pull requests, design docs, experiment logs and Slack threads. Patent Feasibility learns how your company builds, then mines every system and database you connect to find patentable innovation, draft the disclosure, and route it to counsel before a public release starts the clock.

Request Access → See Candidate Inventions
Candidate Inventions · last 90 days Sample · illustrative data
CandidateNoveltyValueRecommend
Speculative decoding with learned draft-model routing14 PRs · 3 design docs · 2 W&B runs · 4 inventors
HighHighFile patent
Synthetic-data dedup via perceptual hashing of embeddings6 PRs · 1 RFC · Slack #data-eng
MediumHighFile patent
Customer-specific RLHF reward weighting recipeNotebook · internal wiki · 2 inventors
HighMediumTrade secret
GPU-aware batch scheduler for mixed-length requests9 PRs · conference talk draft ⚠ disclosure in 21 days
MediumMediumUrgent review

Step 1 · Understand the company

It starts with how you build — not a keyword list

Patent Feasibility first maps your R&D process: which teams own which systems, how ideas move from Slack to design doc to pull request to experiment to release, and where your product differs from the state of the art. That map tells it where innovation is likely to surface — and what counts as routine work for your team.

GitHub · GitLab · code, PRs, reviews Jira · Linear · tickets & epics Confluence · Notion · design docs, RFCs Slack · engineering channels Google Drive · specs & decks MLflow · Weights & Biases · experiments Jupyter · research notebooks Data warehouses · benchmark results

Step 2 · Mine every system

Signals that an invention just happened

Claude reads across your connected sources looking for the patterns that usually mark a technical advance — then groups related evidence from many systems into a single candidate invention.

📈

Measured Improvements

Experiment runs and benchmarks that beat the previous baseline on speed, memory, accuracy or cost — the concrete technical improvement that eligible claims are built on.

🏗️

New Architectures & Algorithms

Design docs and pull requests that introduce new components, data flows or training methods rather than routine refactors or dependency bumps.

🧵

Problem-Solution Threads

Slack and Jira discussions where a hard technical problem was stated, alternatives were rejected, and a non-obvious solution was chosen — exactly the story a patent tells.

🔗

Cross-System Clustering

A PR, an RFC, three experiment runs and a Slack thread about the same idea become one candidate, with every source linked as evidence.


Step 3 · Triage & act

From raw signal to counsel-ready disclosure

⚖️

Patent or Trade Secret?

Not every invention should be patented. Methods that can't be observed in your product are often better kept as trade secrets; detectable, licensable features usually belong in a patent. Each candidate gets a recommendation with the reasons.

📝

Auto-Drafted Invention Disclosure

Problem, prior approaches, the solution, measured results and possible variations — drafted from your own sources, ready for the inventors to correct and sign.

👥

Inventor Attribution

Proposes inventors from commit, review and document authorship, separating conception from implementation so counsel can make the final inventorship call.

⏳

Disclosure-Deadline Alerts

Watches for planned talks, preprints, open-source releases and product launches. A public disclosure starts the one-year U.S. grace period and can destroy novelty outright in absolute-novelty jurisdictions such as Europe; China's grace period is six months and narrow. Each candidate shows its clock per jurisdiction.

🔎

Prior-Art Pre-Check

Every candidate can be sent to Patent Quality with one click for a patentability score and prior-art search before counsel spends an hour on it.

🔐

Confidential by Design

Read-only connectors, tenant-isolated processing and an access log of every source viewed. Your sources are never used to train models.

Patent Feasibility identifies candidate inventions for review by your team and patent counsel. Recommendations are decision support, not legal advice.


Pilot Program

Find the inventions you didn't know you had.

Connect two sources and get a candidate-invention report from your last 90 days of work. Design partners get direct access to the founding team.

Start a Pilot → See Patent Quality →