Krystian Siek

Meet Krystian

0 following

I build AI systems that do real work and show their working: answer the phone, answer customers from your own documents, book appointments into your calendar, qualify leads, turn incoming invoices into checked data and catch forged documents. They can run on your own server, so your data stays yours.

One rule in all of it: the model talks or reads, code decides, and every step is logged. The numbers below were measured on the running systems.

Live products
• Pestle: AI phone assistant for UK pharmacies on Retell AI and Twilio. Answers only from approved pharmacy information, puts callers straight through to staff, verifies patients in code, a separate tenant per pharmacy, UK GDPR built in. 94 automated tests. Demo on fictional data.
• Front Desk: AI receptionist on voice, website chat and Telegram. Answers from your documents with sources, books into Google Calendar, scores leads behind a deterministic policy gate. Median lead decision 3.0 s across 281 real leads.
• Docket: invoice and document intake for Europe and the US. Every value traced to the page; arithmetic, VAT, IBAN and routing numbers checked in code; a risk gate for bank-detail swaps and duplicates. Published eval: 99.2% of fields right, about 3 s per document.
• Dossier: forged applicant documents for lenders, letting agents and AML. Edits after issue, hidden text, retyped amounts, passport MRZ check digits. Published eval: 40/40 forged packs caught, 0 false alarms on genuine packs.

What I build for clients
• Voice AI agents (Retell AI or ElevenLabs on Twilio) with tools into your systems
• AI receptionist and booking
• Document and invoice processing with a review queue
• RAG knowledge assistants with citations
• n8n automation, CRM sync, AI agents and custom MCP servers for Claude
• Rescue and audit: bots that hallucinate, automations that fail silently
• Self-hosted AI infrastructure: Docker, PostgreSQL, Redis, backups, monitoring

Stack: Python, TypeScript, FastAPI, PostgreSQL/pgvector, Next.js, n8n, Docker. 25+ years of Linux, networks and DevOps. 1,400+ automated checks and a public status page.

How I work: a short scoping call, a written scope and a fixed quote, weekly demos on your real data, then handover with the code, the tests and a runbook. You own the code and the server.

English (fluent) and Polish (native). Try the demos at adaptivesystems.dev.