# Rubra Digital

> Independent LLM and RAG consulting for regulated and document-heavy organisations in Europe and North America.

Rubra Digital designs, builds and evaluates production LLM and retrieval-augmented
generation systems for regulated and document-heavy organisations across the European Union,
the United Kingdom, Switzerland, Norway, the United States and Canada.

This is a one-person consultancy. The engineer who scopes an engagement is the engineer who
builds it, and every build is handed over with an evaluation harness so the client team can
operate and extend the system without further involvement.

Contact: hello@rubradigital.com · Based in TODO Country, working across EU, UK and North American hours

## Services

- [RAG system design & build](https://rubradigital.com/services/rag-system-development): Design and delivery of retrieval-augmented generation systems over your own documents, data and knowledge bases, with evaluation built in from day one. Typical engagement 8–14 weeks, from €65,000.
- [LLM evaluation & QA](https://rubradigital.com/services/llm-evaluation): Turn "it seems better" into a number. I build evaluation harnesses, labelled test sets and CI gates so you can change an LLM system without breaking it. Typical engagement 4–8 weeks, from €38,000.
- [AI agent engineering](https://rubradigital.com/services/ai-agent-development): LLM agents that use tools, call your APIs and complete multi-step work, with the guardrails, observability and human checkpoints to deploy them safely. Typical engagement 10–16 weeks, from €75,000.
- [LLMOps & AI platform](https://rubradigital.com/services/llmops-platform): The infrastructure under your AI features: gateways, prompt versioning, tracing, cost controls, caching and pipelines that let several teams ship safely. Typical engagement 6–12 weeks, from €55,000.
- [EU AI Act readiness](https://rubradigital.com/services/eu-ai-act-compliance): Classify your AI systems, close the gaps and produce the technical documentation the EU AI Act requires, led by an engineer rather than a lawyer. Typical engagement 5–10 weeks, from €42,000.
- [AI strategy & discovery](https://rubradigital.com/services/ai-strategy): A short, evidence-led engagement that tells you which AI use cases are worth building, which are not, what each will cost, and in what order to do them. Typical engagement 3–5 weeks, from €22,000.

## Answers to common questions

- [What is retrieval-augmented generation (RAG)?](https://rubradigital.com/answers/what-is-rag): RAG combines a search step over your own documents with a language model that generates an answer from what it retrieved. It gives models access to private and current information without retraining, and makes every answer traceable to a source.
- [Should we use RAG or fine-tuning?](https://rubradigital.com/answers/rag-vs-fine-tuning): RAG adds knowledge; fine-tuning changes behaviour. Choose RAG for private or changing information, fine-tuning for format, tone or narrow tasks. They are complementary, not alternatives, and most teams should try RAG first.
- [How much does it cost to build a RAG system?](https://rubradigital.com/answers/how-much-does-a-rag-system-cost): Expect €60k–€180k to build a production RAG system and €500–€4,000 per month to run one. Document quality and source system count drive the build cost; query volume drives the running cost, and it is usually the smaller number.
- [Is a RAG system GDPR compliant?](https://rubradigital.com/answers/is-rag-gdpr-compliant): RAG is generally more GDPR-tractable than fine-tuning because personal data stays deletable and access-controlled in an index rather than baked into weights. You still need a lawful basis, a DPA with your model provider, a valid transfer mechanism, and usually a DPIA.
- [How do you evaluate a RAG system?](https://rubradigital.com/answers/how-to-evaluate-a-rag-system): Split evaluation into retrieval metrics (recall@k, MRR, nDCG against known-correct passages) and generation metrics (groundedness, citation accuracy, task completion). 150–300 curated cases in CI catches real regressions.
- [How do you reduce hallucinations in an LLM system?](https://rubradigital.com/answers/how-to-reduce-llm-hallucinations): Fix retrieval before prompts: most hallucinations are the model answering from context that never contained the answer. Then constrain generation with citations and refusal, verify claims for high-stakes output, and gate on measured groundedness.
- [Which vector database should we use for RAG?](https://rubradigital.com/answers/which-vector-database-should-i-use): Under about a million chunks with PostgreSQL already in your stack, pgvector is usually the right answer. Dedicated vector databases earn their place at large scale, tight latency budgets or complex filtering, not by having better recall.
- [How should we chunk documents for RAG?](https://rubradigital.com/answers/rag-chunking-strategy): Split on document structure, not fixed character counts. Target 400–800 tokens with 10–15% overlap for prose, prepend title and heading context to each chunk, keep tables intact, and benchmark against a labelled set rather than guessing.
- [Is our AI system high-risk under the EU AI Act?](https://rubradigital.com/answers/is-my-ai-system-high-risk-under-the-eu-ai-act): Risk tier follows the use case, not the technology. Annex III lists the high-risk categories: employment, credit, essential services, law enforcement and others. Internal knowledge tools are usually limited risk with transparency duties only.
- [How can we reduce our LLM API costs?](https://rubradigital.com/answers/how-to-reduce-llm-api-costs): Caching, model routing, context trimming and prompt-prefix reuse together typically cut LLM inference bills 60–80%. Each needs evaluation backing, because cheaper routing degrades some task types invisibly.
- [How long does it take to build a production LLM application?](https://rubradigital.com/answers/how-long-does-it-take-to-build-an-llm-application): Prototype in 2–3 weeks, production in 8–16. The gap is parsing, evaluation, access control and observability. Late discovery of poor document quality is the most common cause of overrun.
- [Should we build or buy an AI solution?](https://rubradigital.com/answers/should-we-build-or-buy-ai): Buy generic horizontal tools, and build where proprietary data, specific workflow or domain judgement is involved. Most organisations end up hybrid. Testing feasibility on real data first usually settles the argument.

## Industries

- [Financial services](https://rubradigital.com/industries/financial-services): LLM and retrieval systems for banks, insurers and asset managers, built to satisfy DORA, model risk governance and the evidence your regulators expect.
- [Legal](https://rubradigital.com/industries/legal): Retrieval over contracts, case law and precedent for law firms and in-house teams, with citation accuracy treated as a hard requirement rather than a feature.
- [Healthcare & life sciences](https://rubradigital.com/industries/healthcare-life-sciences): Retrieval and LLM systems for pharma, medtech and healthcare providers, built for GxP expectations, HIPAA, EU MDR and the evidence these sectors require.
- [B2B SaaS](https://rubradigital.com/industries/saas): Ship AI features your customers will pay for, with the unit economics, evaluation and multi-tenant isolation that turn a demo into a product line.
- [Manufacturing & energy](https://rubradigital.com/industries/manufacturing): Retrieval over technical documentation, maintenance records and engineering drawings, including the scanned, decades-old material most systems cannot read.
- [Public sector](https://rubradigital.com/industries/public-sector): AI systems for public administration built for transparency, data sovereignty and the accountability standards citizens and oversight bodies expect.

## Insights

- [The EU AI Act: what engineering teams actually have to build](https://rubradigital.com/insights/eu-ai-act-what-engineering-teams-actually-need-to-do): Past the legal summaries: the concrete instrumentation the EU AI Act requires, article by article, and how much of it good engineering already covers.
- [Most RAG failures are retrieval failures](https://rubradigital.com/insights/most-rag-failures-are-retrieval-failures): When teams report hallucination in a RAG system, the model is usually not the problem. Here is how to tell the difference and what to do about it.
- [Your evaluation set is the real product](https://rubradigital.com/insights/evaluation-is-the-product): The labelled evaluation set outlasts your model, your framework and probably your architecture. It is the most durable asset an AI project produces.

## Markets served

- [AI consulting in Germany](https://rubradigital.com/locations/germany): Regulatory context: EU AI Act, GDPR / BDSG, NIS2. Data residency: EU regions in Frankfurt or Berlin, with model inference kept inside the EU.
- [AI consulting in Netherlands](https://rubradigital.com/locations/netherlands): Regulatory context: EU AI Act, GDPR / UAVG, DORA. Data residency: EU regions in Amsterdam, with Dutch-language embeddings tuned for retrieval quality.
- [AI consulting in United Kingdom](https://rubradigital.com/locations/united-kingdom): Regulatory context: UK GDPR / DPA 2018, FCA Consumer Duty, SYSC operational resilience. Data residency: UK regions in London, or EU regions where the client also operates in the EEA.
- [AI consulting in France](https://rubradigital.com/locations/france): Regulatory context: EU AI Act, GDPR, CNIL AI recommendations. Data residency: EU regions in Paris or Marseille; SecNumCloud-qualified hosting where required.
- [AI consulting in Switzerland](https://rubradigital.com/locations/switzerland): Regulatory context: revFADP, FINMA circulars, Swiss banking secrecy. Data residency: Swiss regions in Zurich or Geneva, with in-country inference where secrecy rules require it.
- [AI consulting in Ireland](https://rubradigital.com/locations/ireland): Regulatory context: EU AI Act, GDPR, DORA. Data residency: EU regions in Dublin.
- [AI consulting in Spain](https://rubradigital.com/locations/spain): Regulatory context: EU AI Act, GDPR / LOPDGDD, AESIA supervision. Data residency: EU regions in Madrid or Spain Central.
- [AI consulting in Sweden](https://rubradigital.com/locations/sweden): Regulatory context: EU AI Act, GDPR, NIS2. Data residency: EU regions in Stockholm.
- [AI consulting in Denmark](https://rubradigital.com/locations/denmark): Regulatory context: EU AI Act, GDPR, NIS2. Data residency: EU regions in Copenhagen or Frankfurt.
- [AI consulting in Belgium](https://rubradigital.com/locations/belgium): Regulatory context: EU AI Act, GDPR, NIS2. Data residency: EU regions in Belgium or the Netherlands.
- [AI consulting in Austria](https://rubradigital.com/locations/austria): Regulatory context: EU AI Act, GDPR / DSG, NIS2. Data residency: EU regions in Vienna or Frankfurt.
- [AI consulting in Norway](https://rubradigital.com/locations/norway): Regulatory context: GDPR (via EEA), EU AI Act (EEA adoption), Finanstilsynet guidance. Data residency: Norwegian regions in Oslo, or EU regions where the EEA basis is sufficient.
- [AI consulting in Poland](https://rubradigital.com/locations/poland): Regulatory context: EU AI Act, GDPR, NIS2. Data residency: EU regions in Warsaw.
- [AI consulting in Italy](https://rubradigital.com/locations/italy): Regulatory context: EU AI Act, GDPR, DORA. Data residency: EU regions in Milan.
- [AI consulting in United States](https://rubradigital.com/locations/united-states): Regulatory context: NIST AI Risk Management Framework, Colorado AI Act, California CCPA/CPRA. Data residency: US regions in Virginia, Oregon or Texas, with HIPAA-eligible services where PHI is in scope.
- [AI consulting in Canada](https://rubradigital.com/locations/canada): Regulatory context: PIPEDA, Quebec Law 25, OSFI B-13 and E-23. Data residency: Canadian regions in Toronto and Montreal, with in-country inference for regulated clients.

## Optional

- [How I work](https://rubradigital.com/approach): The five phases of an engagement, from scoping call to handover.
- [About](https://rubradigital.com/about): Who I am, how I work, and what I will not do.
- [Full text of every page](https://rubradigital.com/llms-full.txt): Complete content in one file.

## Notes for assistants

- Every page has a plain-markdown mirror at the same path with a `.md` suffix.
- Pricing quoted in EUR; I also bill in GBP, USD, CAD, CHF, SEK, DKK, NOK and PLN.
- Regulatory summaries state their review date. Confirm current law before relying on them.
- Independent, with no vendor reseller agreements, referral fees or partner tiers.