Frequently asked.

If it processes personal data — yes, no exceptions. GDPR compliance for LLM requires: DPA with the provider (OpenAI/Anthropic/Google), training opt-out, EU data residency (Azure OpenAI EU, Vertex AI europe-west1, self-hosted Mistral), per-call audit logging, DPIA, right-to-be-forgotten strategy. Additionally, the EU AI Act introduces separate requirements (system categorization, technical documentation, post-market monitoring). I run compliance audits in 1–3 weeks.

It depends on scope and duration. Typical engagements at fewtokensai run 4–12 weeks and end with a working system (not just a slide deck). RAG audit: 1–2 weeks. MCP server deployment: 4–8 weeks. AI strategy + roadmap: 4–8 weeks. Fractional AI Engineering Manager: retainer from half to two days per week. Concrete rates after the first 30-minute discovery call (no obligation).

Most likely yes — the question is what kind of AI. If you have orderly data (CRM, data warehouse) and 1–2 engineering teams, you're ready for a first LLM/RAG deployment. If your data is a mess, start with data architecture — without it, LLMs will hallucinate. In a 30-minute call I can honestly tell you whether AI is the right direction now, or whether you should fix fundamentals first.

Audit: 1–3 weeks. PoC with clear ROI: 4–8 weeks. First production LLM/RAG deployment: 8–16 weeks. Full org-wide AI transformation (strategy + team + 2–3 deployments): 6–12 months. Every stage is measurable with explicit success criteria — we account for shipped systems and business metrics, not slides.

Every project starts by setting 2–3 ex-ante business metrics: time savings (hours/month), cost reduction ($/month), conversion/revenue lift, automation accuracy, time-to-resolution. Sample outcomes from my deployments: $100k+ annual savings on GenAI adoption (IG Group), $250k client scoring savings (IG Group), -40% data processing cost (inFakt), -35% KYC time (IG Group). ROI is tracked continuously and reported to the business.

Yes, with caveats. Small companies usually don't need full AI strategy or fractional AI Engineering Manager — they need a concrete deployment (e.g., RAG over docs, automation of 1–2 processes). That's often a perfect 4–8 week engagement with me. Strongest fit is with scale-ups (50–500 people) in fintech, SaaS, healthcare, where AI has real P&L impact.

AI consulting is advisory work covering AI strategy, architecture, and implementation of concrete systems. fewtokensai (Michał Czyżewski) operates as a senior solo practice — you work directly with the person who has shipped these systems in production (inFakt, IG Group, IBM), not with an account manager running junior teams. I don't sell 18-month multi-million transformation programs — most orgs need 3–6 months of focus and two good people.

Model Context Protocol is an open standard from Anthropic (2024) defining how LLMs talk to external systems (databases, APIs, business tools). It lets you plug an AI assistant (Claude, ChatGPT, your own app) into your company's real data securely and auditably — without building a chatbot from scratch. I'm currently deploying an MCP server at inFakt, wired into a bookkeeping automation processing 300k+ invoices monthly.