How we started
Omoikane was built because someone close to me — a thoughtful, experienced operator with a track record in international development and the cultural sector — was applying to twelve jobs a week and getting nothing back.
The problem wasn’t that her CV was bad. Her CV was good. The problem was that she was spending eight hours per application getting it good — translating her experience into the right register, finding the matching evidence, drafting a letter that didn’t sound desperate, double-checking the company against her values. The eight hours were correct. They produce strong applications. But they don’t scale to a real job search, and at that volume they consume the energy that the actual hiring conversations need.
So this is the workspace that does the slow translation work — accurately, with full citation, in your voice — so the human stays human-shaped for the parts of the search only humans can do.
The name comes from 思兼神 (Omoikane), the kami of wisdom and deliberation in Japanese mythology. The kami who is asked to think before the gods act. The product’s whole behaviour is in the name.
Who we are
Omoikane.Coach is a small two-person operation.
Elli.Z.G. is the legal owner and the public face — content, copy, UI, UX. The product feels the way it feels because Elli designs every screen and iterates from there. Omoikane.Coach is registered as a Dutch eenmanszaak (sole proprietorship) in Leiden, the Netherlands, in Elli’s name; she is responsible for customer relationships, data handling, and product decisions. LinkedIn · withelli.com
Elli leads what the user sees and how it speaks; the engineering underneath — the Rust application, the infrastructure, the AI-dispatch and safety layers — is built and maintained in-house. Not a launch. Not a Series A. A tool, maintained.
Under Article 14 of the EU AI Act, Elli is Omoikane’s designated AI Oversight Officer — accountable for the human oversight of the AI system. Technical and model-level oversight is carried out by the in-house engineering.
What’s under the hood:
- Hosted in Norway, witnessed in Leiden. Two Norwegian database hosts carry your data, synchronously replicated; a Leiden host casts the cluster’s tie-breaking vote and holds no user data. EEA-only, end to end.
- Your key, your model. On the bring-your-own-model tier, drafting runs on whichever LLM provider you configure. The managed tiers — not yet open — are planned to run through Mistral (EU); the full list lives in the sub-processor register. Every draft is built from your own evidence: each bullet carries a source reference to a document you uploaded, unsupported requirements are surfaced as gaps, and drafts without citations get flagged.
- Core infrastructure is self-hosted. Authentication, storage, search index, observability, vector database. The complete list of third parties that ever touch data is published in the sub-processor register. No third-party analytics or tracking — aggregate page-views come from our own cookieless, self-hosted Umami instance (see the cookie policy). No AI-training partners.
- Open-source where possible. All infrastructure components are FOSS-licensed — Authentik, YugabyteDB, Qdrant, SeaweedFS, LiteLLM, OpenObserve, HAProxy, Reactive Resume. We pay for what we use; we don’t depend on companies that could disappear.
If you find it useful, the best thing you can do is use it carefully. If it doesn’t fit your situation, the second-best thing is to tell us why.
The legal entity, registration numbers, and contact channels are documented on the Imprint page.