A self-improving engine that takes an idea, grounds it in 791 real patents and 29,175 verified atoms, generates hundreds of mechanism variants, compresses the best into novel hybrids, and exits with a complete attorney-ready patent package — and 31 commercial documents — before a human has had coffee. Then it turns itself inward and improves.
Who this is for
Accelerate prior-art synthesis and draft generation. Attorney-ready packages go directly to prosecution — no reformatting required.
Turn mechanism concepts into commercial IP pathways and 90-day market entry blueprints across active domains.
Evaluate a self-improving IP refinery, not a single patent. The moat is the substrate, the provenance, and the compounding loop.
Convert frontier technical insights into protectable assets. The system generates, rates, and packages without human coordination.
Pieta is a fully autonomous IP generation engine that turns a single anchor mechanism into attorney-ready patent packages — grounded in 791 real EPO patents and 29,175 verified mechanism fragments, with zero hallucination and full claim traceability.
Over 18 months, it evolved from a first documented run of 3,457 candidates into a 6-phase pipeline spanning Hunter-Killer ideation, E1.5 sectional trimming, SYNTH hybrid compression, and a 31-document commercial strategy layer — all without external headcount.
The system has delivered 67+ attorney-ready patent packages across three active IP domains — 55 from the umbrella IP Foundry proof (Tests 1 & 2), 7 from the cooling domain, 5 from metsigen — and then turned its own combinatorial engine inward: Pieta Mirror generated 7,490 self-analysis ideas and identified 15 buildable system improvements, with 7 independent derivations converging on the same structural fix.
Total all-time cost: $173.88. Equivalent manual IP foundry engagement: $300,000+. The engine that files is itself the IP: PPN00002480.
The system is live, production-grade, and self-improving. It is seeking the right institutional partnership for its next phase — whether that is IP filing and commercialisation, joint research, or strategic alignment. A 30-minute walk-through of any section of this brief is available on request.
IP generation is broken — too slow, too expensive, and industrially blind. The countries that file fastest set the standards. The companies that file first own the markets.
Global patent filing is accelerating. Volume alone no longer differentiates strong portfolios: the institutions that close the quality and speed gap — filing grounded, novel mechanisms faster than competitors can prior-art them — define the next decade of technical standards. Speed and precision are the only remaining leverage.
Noise — 3.2M utility models buried in prior art. Speed — provisional windows close while lawyers draft. Blind spots — no system sees structural isomorphisms across domains. Hallucination — LLMs generate claims with no grounding.
$23.7B+ in documented losses attributable to missed provisional windows, failed freedom-to-operate searches, and standard-essential patent traps. One missed filing can lock a company out of a market permanently.
A manual IP foundry engagement — attorney team, technical consultancy, 18-month runway — costs $300,000+ for a comparable patent portfolio. Pieta delivered 67+ attorney-ready packages across three domains for $173.88 all-time. Zero external headcount. Under 30 days from anchor definition to first filing-grade output.
Before a single patent idea could be generated, the engine needed something no LLM has: a grounded substrate. 791 real EPO patents across 32 domains, 7 meta-families — parsed, atom-extracted, and verified against a multi-model Reality Gate before a single claim was written.
Draw → Triangulate → Diverge → Ground → Audit. Five independent passes against the atom pool. Every output is traceable to its source atoms and the original patent text. No claim exists without provenance.
Three atom pools, built and verified over multiple runs. Legacy pool (first extraction pass), Universal Library corpus, and Synthetic verified pool — all passed the Reality Gate before entering production.
Gen 2 scaled run (2026-07-17): 142 synthetic atoms through the full Reality Gate under concurrent load. The engine held. No hallucinated atoms entered the pool.
334 chapters · 22 theses · 103 blueprints
$57.16 last 30 days · 25,680 API calls · 124M tokens
Gen 2 scaled run — atoms that passed the Reality Gate
In the production pool — every atom earned its place
The refinery proved the engine. The Hunt armed it. Instead of exploring the full idea space randomly, Pieta is given a specific mechanism target — the Anchor — and the grid systematically explores every way atoms from the substrate can combine with it to produce something new and fileable.
Each IP project starts with a single high-value mechanism target — a specific engineering challenge in a domain where IP whitespace exists. The anchor defines the exploration boundary: every idea the grid produces must touch this mechanism.
The grid explores each anchor section's mechanism space systematically, producing forge ideas the operator wouldn't have considered independently. 50–200 forge ideas per run, each with a forge code, anchor section binding, and IPS pre-score on creation.
Every forge idea passes through a 4-model evaluation panel: three independent DeepSeek calls plus one GLM-4. Each model scores independently across IPS dimensions — novelty, non-obviousness, claim breadth, prior-art clearance, enablement. The panel reaches consensus. Diamond / Gold / Silver.
11,076 ideas registered across all projects and source tiers — Forge (HK grid), SYNTH (hybrids), T3 (gap critic pool), P1 (anchor combo pre-screening), EXT (external injections). Diamond 22 · Gold 64 · Silver 157 · Bronze 10,833. The Silver and Bronze pool is not waste — it is the recombination substrate for the next run.
The first full end-to-end IP Foundry runs — grounded against 791 real EPO patents, scored under filing-grade weights. Test 1 established the pipeline works. Test 2 introduced cross-day fusion and delivered every filing-grade idea as an attorney-ready patent draft.
Three generation tracks harvested from one day, fused directly, and scored against 791 EPO prior-art patents. Established that the foundry works end-to-end and that idea volume ≠ top quality.
3,457 ideas in88 super-patents27 filing-gradeIPS 79 top4,596 candidates across two days distilled into a 377-idea cross-day pool. Fused under filing weights, auto-fault-checked, then turned into lineage-traced attorney-ready patent packages.
4,596 → 377 pool117 super-patents28 filing-gradeIPS 83 topAttorney-ready patent drafts across both tests
Raised 79 → 83 by cross-day fusion in Test 2
Test 2 vs Test 1 — more super-patents, deeper Silver pool
Every idea grounded against real EPO patents at scoring time
Claims JSON · Claims MD · Technical SVG patent drawings · Commercial go-to-market kit · Prior art search with cited references · Full application draft (markdown) · Application draft (.docx, attorney handoff format). 0 manual steps between idea and completed package.
From the foundry proof, three focused IP projects were built and run — each with its own anchor mechanism, anchor sections, and full forge → E1.5 → SYNTH → Patent pipeline. The same autonomous stack, aimed at three different engineering problems.
PHD — the Pigtail Annular-Weir Cartridge. Separates pneumatic and hydraulic energy paths inside a single actuator body — eliminating the cross-coupling that limits precision in hybrid systems.
Adaptive Nunome Cold Plate — liquid cooling for AI compute units exceeding 10 kW. Two 8-hour autonomous runs over 2 days. One DIAMOND-rated claim combining mechanisms the forge couldn't produce in a single idea.
The original IP Foundry runs — 8 physical domains, cross-day fusion across the full candidate pool. Tests 1 & 2 established the filing-grade threshold and delivered the first complete attorney-ready packages.
Every forge idea enters a multi-stage compression that eliminates weak mechanism sections and combines the strongest surviving mechanisms into novel hybrids. The hybrids — not the forge ideas — are what gets filed.
All ideas registered across all projects and tiers
IPS ≥ 75 across all projects — priority filing candidates
IPS 70–74 — filing-grade, queued for patent branch
Attorney-ready drafts across all 3 active domains (55 umbrella + 7 cooling + 5 metsigen)
The substrate, the Genome DB, and all three active IP domains are live. Leave your email and we'll respond directly — or arrange a walkthrough at a time that suits you.
The IP pipeline is one track of a larger system. Every idea that clears the patent gate also generates a full commercial strategy package — and a 12-stage industrial cascade that maps it from raw resource through to institutional power graphs and 90-day market entry blueprints.
Every qualified idea is traced from raw resource streams through industries, processes, components, products, and markets — all the way to institutional power graphs, VCP+MCDA rankings, four strategic narrative perspectives, and an adversarially-critiqued 90-Day Blueprint. 17 node types, 18 edge types. Output feeds NEXUS episodic memory for cross-idea pattern matching.
Idea to finished artifact — no human coordination required
Full strategy layer output per qualifying idea
Resource → Institutional Power → 90-Day Blueprint
Between idea input and finished deliverable package
The same combinatorial topology engine that generates patentable inventions — turned inward. Instead of finding structural isomorphisms between patents, it finds structural isomorphisms between system components and their failure modes. The result: Pieta's own architecture becomes the next anchor.
Pieta Mirror is a git worktree on branch lab/mirror — an isolated copy of the live system where proposed improvements are tested before they touch production. The same engine that hunts patents hunts its own weaknesses.
Every AI system has a blind spot: the people who built it cannot see its structural flaws from inside. Pieta solved this by treating its own architecture as the domain and running the full HK grid against it. The improvements it found are the ones that couldn't be seen from inside.
When seven independent derivations from different starting atoms all identify the same structural weakness, that is not random — that is the engine repeatedly finding a real gap from different directions.
Generated by Pieta about its own architecture
Identified and scoped — with time estimates
Found from multiple independent angles
Seven independent derivations to the same structural fix
A system that can improve itself is a system that compounds. The 15 Mirror improvements apply directly to the next production run — raising IPS ceilings, fixing prior-art density errors, and unlocking the 889 ideas currently hard-blocked by the seven-origin finding. That is what the commercial products below deliver on top of.
Two products. One sells the engine's outputs — packaged intelligence for IP teams. The other runs the engine on your domain and gives you the IP.
A VP IP or CTO subscribes to domain-specific patent intelligence: prior-art landscapes, whitespace maps, filing-grade idea feeds, and competitive freedom-to-operate assessments — all grounded in the atom substrate, not LLM guesswork.
Run the full pipeline on a client's anchor mechanism. Pieta generates the ideas, rates them, runs E1.5 + SYNTH + Patent Branch, and delivers a portfolio of attorney-ready packages. Client pays royalties on licensed IP. Pieta retains 100% of output IP generated from its own substrate.
| Capability | Pieta | PatSnap | Palantir | Google Patents AI | Pure LLMs |
|---|---|---|---|---|---|
| Grounded substrate (real patents) | |||||
| Autonomous idea generation | |||||
| Attorney-ready patent drafts | |||||
| Commercial strategy automation | |||||
| Self-improving architecture |
Manual IP Foundry engagement: $300,000 · 18 months for a comparable patent portfolio (law firm + consultancy). Pieta equivalent: $173.88 all-time · under 30 days from anchor definition to first attorney-ready packages. Scope delivered: 67+ packages, 31 commercial documents per qualifying idea.
The moat is not the technology — it is the substrate, the provenance, and the compounding. Every new atom run makes the next idea run better. Every new idea run adds to the substrate.
29,175 atoms — each traceable to its source patent and its Reality Gate verdict. Competitors start with text. Pieta starts with verified mechanism fragments.
Every claim in every patent package traces back to the atoms that generated it, and the patent text those atoms came from. Zero black boxes between substrate and output.
Every SYNTH run seeds the next. 41 of 52 cooling hybrids are in the survivor pool — the next run starts with proven mechanisms, not blank ideas.
PPN00002480 — the substrate construction and the grounding methodology itself are IP. The engine that makes IP is itself IP.
The engine audits its own architecture using the same grounding methodology. Improvements are tested in the Mirror worktree before production — no other IP platform has this.
11,076 ideas registered with scoring history. The provenance record is an asset — it proves the system's calibration history, not just its current output.
The original pre-seed ask was to file 3–5 provisional mechanisms. We built a system that generates, evaluates, compresses, and attorneys-ready-packages at scale. Here is what we have and what we need next.
Starting from: "file 3–5 provisional filings." Ending with: a fully autonomous IP generation pipeline, three active domains, 67+ attorney-ready packages, a self-improving mirror, and $173.88 all-time spend.
Three near-term milestones that convert the built system into filed IP and first licensing conversations.
I am the only person who has operated every pipeline stage end-to-end. I designed the safety architecture. I filed the patent. There are no co-founders, no equity splits, no team leaks. The IP chain is clean.
Now: File remaining P1 Diamond patent branches. Run Mirror improvements (density-normalised thresholds — 889 ideas currently hard-blocked by this one fix). Land first licensing conversation in the cooling domain.
Q1 2027: Physics validation layer — atoms grounded against simulation data in addition to patent text. Raises the IPS ceiling and reduces prosecution risk on the strongest claims.
2027: 1,500+ patent packages across 6+ domains. Licensing revenue from cooling and metsigen domains. Substrate reaches full universal coverage across the 7 meta-families.
Request a 30-minute walkthrough — see the engine, the output structure, and the commercial logic behind Pieta. The substrate, the genome DB, and the pipeline are all live.
If you want to follow Pieta as it evolves — new domains, Mirror improvements, licensing milestones — leave your email and we'll send major updates only. No noise.