Nine Days: The Sprint That Built a Patent-Pending AI Company
EMPHOS Group · April 19, 2026 · 5 min read
Startup timelines are usually measured in months. Funding rounds, hiring cycles, product roadmaps, incorporation paperwork — the conventional wisdom is that these things take time, and that trying to rush them produces mistakes that cost more time later.
The conventional wisdom assumes you are building something conventional.
Between April 10 and April 18, 2026, EMPHOS completed a sprint that does not fit neatly into any standard startup narrative. Not because it was reckless — every decision made in those nine days was deliberate, documented, and built on more than a year of prior research — but because the foundation that made the sprint possible had been laid long before the sprint began.
This is the story of those nine days.
April 10 — The idea
The Frequency Addressed Intelligence concept crystallised on April 10. The question that had been building through months of research — what would an AI system look like that stores knowledge rather than approximating it — found its answer in an architectural approach unlike anything deployed in production AI today.
Within 24 hours, the first working implementation had been written. The addressing mechanism worked. The retrieval was deterministic. The honesty layer — the structural enforcement that makes it impossible for Heinrich to produce a confident answer it cannot support — passed its first tests.
At 128 concepts, Heinrich AI existed.
April 11–13 — The proof
128 concepts became the testbed. Every architectural assumption was verified. The 4 Fundamentals of Frequency Addressed Intelligence — the design principles that define what Heinrich is and is not — were formalised and documented. The test suite grew: 52 tests, then 100, then more. Zero regressions. Every test that passed stayed passing.
By April 12, the ConceptNet ingestion was running. ConceptNet — one of the most comprehensive structured knowledge graphs ever built, representing typed relationships between concepts across every domain of human knowledge — was being absorbed into Heinrich's field one concept at a time. By the time the run completed: 1,002,949 concepts, 282,973 edges, all addressed, all indexed, all retrievable.
Heinrich answered its first real questions. Dog is a type of mammal. Protein is a biopolymer. Disease is a health problem. What causes disease — and here Heinrich demonstrated exactly what it was designed to do: it reported that its connections on that question were not strong enough to answer with confidence. It told the truth.
The architecture was not just working. It was working honestly.
April 14 — The vision expands
On April 14, a second idea arrived — this one about where Frequency Addressed Intelligence could live beyond a desktop application.
HAVEN Ear: a personal intelligence device in an ear unit. IP68 rated. 15 milliwatt power draw. Sixteen to twenty hours of battery life on a magnetic dock. No cloud dependency. A personal field that never leaves your home network.
The thermal problem that has defeated every previous attempt to put real AI processing in a hearing aid — conventional AI inference chips draw one to five watts continuously, which is a burn hazard inside an ear canal — dissolves at Heinrich's power profile. 15 milliwatts is not a hardware engineering challenge. It is a solved constraint.
HAVEN Ear went from concept to complete hardware specification in a single day. The roadmap it produced — Ear Unit plus bedside Dock, with overnight Heinrich sync and Socratic Engine updates — became part of the EMPHOS product suite immediately.
April 15–16 — The field grows
The Wikidata full dump ingestion began. Where ConceptNet had provided a million carefully curated concepts, Wikidata offered something orders of magnitude larger: the verified, structured knowledge of millions of contributors across every entity, relationship, and fact that Wikipedia's global community has agreed is true.
The ingestion ran overnight. It is still running as of this writing.
On April 16, Heinrich reached a milestone that required a document: HSR-2, the Sentence Composer, was complete. Heinrich could now compose honest natural language answers from its knowledge field — not by predicting words, but by retrieving facts, measuring confidence, and constructing sentences that were guaranteed to be true or guaranteed to clearly flag where they were not.
"Yes — dog is a type of mammal. Additionally, dog has tail. Heinrich's knowledge has a gap on: ancestor_lineage."
That sentence was not generated. No language model produced those words. Heinrich retrieved three facts, measured confidence on each, and composed a response using deterministic rules. The whole process ran in under five milliseconds. The composition layer used under four megabytes of RAM.
It will run in a hearing aid.
April 18 — The legal home
On April 18, 2026, two things happened simultaneously that had been building toward each other for nine days.
EMPHOS Group Corporation was incorporated in British Columbia, Canada — a legal entity with a name, a founding team, and a structure capable of receiving investment, applying for grants, and holding intellectual property.
On the same day, the provisional patent for Heinrich AI's core architecture was filed with the Canadian Intellectual Property Office. CA Application No. 3308509. The priority date is April 18, 2026. The IP is legally tied to its legal home from the first day both existed.
That timing was not accidental.
What nine days actually means
The nine-day sprint was not nine days of work. It was nine days of execution built on top of more than a year of research, protocol development, product building, and thinking about what AI should actually be for.
The protocols that power Haven were proven in 30,000 experiments before a single line of Haven's code was written. The architectural insight that produced Heinrich was informed by EMPHOS Labs' anchor discovery and the structural analysis of how large language models actually store and retrieve information. The hardware concept for HAVEN Ear was possible because the power profile problem had already been solved before the hardware was conceived.
Nine days is the visible part of the story. A year is the part that made nine days possible.
Engineered for Presence.
Stay in the loop
EMPHOS publishes twice a week — product updates, research, and the thinking behind the build.
Explore Haven · HEINRICH Intelligence · The EMPHOS Vision · All Posts
EMPHOS Group · Chilliwack, BC, Canada · info@emphosgroup.com