Low-friction, fixed-price, fixed-time engagements that open the door to larger projects. Each is designed to be delivered in 2–4 weeks and ships with a one-page "what next" proposal — so you can decide what to do with zero pressure.
Your own ChatGPT that reads your documents — and the data never leaves your building.
For: legal, accounting, banks, public sector, any firm with confidential docs.
€5,500 Good · €7,500 Better (SSO + 2 doc sources)
Choose this if your team keeps Googling or paging through your own documents for answers — contracts, policies, case files, tickets — and you want it answered in seconds, with citations, privately.
| What you get | Chat assistant over up to 5,000 of your docs, with citations, on hardware you control. |
| Timeline | 2–3 weeks |
| Success measure | ≥5 demo questions answered correctly with citations from your own docs. |
| What it leads to | A full private knowledge base across more sources and teams, optionally on your own inference hardware under a managed SLA. |
Pick the document you hate rekeying. In 3 weeks it stops being rekeyed.
For: manufacturing, logistics, accounting, banks, insurance, retail ops.
€6,500 Good · €9,000 Better (+ exception UI + validation)
Choose this if staff are manually rekeying the same kind of document every day — invoices, bills of lading, claims, KYC packs — and you can name the single document type that hurts most.
| What you get | Extraction pipeline for one document type (invoices, BoL, claims, contracts, KYC) into CSV/API/ERP, ≥90% field accuracy on a 50-doc sample. |
| Timeline | 3 weeks |
| Success measure | On 50 client-provided docs, ≥90% of target fields extracted correctly. |
| What it leads to | More document types, a validation queue for exceptions, and write-back straight into your ERP or CRM. |
In 2 weeks: a roadmap for private AI — and one model running on your hardware to prove it.
For: banks, public sector, healthcare, regulated industries.
€7,500 Good · €10,000 Better (+ 1 RAG demo + cost model)
Choose this if the board or regulator is asking "what's our AI plan and is it compliant?" — and you want both a defensible answer and proof it runs on your own iron.
| What you get | (a) Readiness report: which workloads fit private AI, TCO vs cloud, compliance posture, roadmap. (b) A working self-hosted LLM endpoint on your hardware serving 1 model via Open WebUI. |
| Timeline | 2 weeks |
| Success measure | Documented assessment + a reachable inference endpoint on client-controlled infra. |
| What it leads to | A full production platform, an AI governance register, and ongoing managed operation under SLA. |
In 2 weeks, your inbox stops being a black hole.
For: any mid-market firm; especially IT service desks, accounting, ops.
€5,000 Good · €7,000 Better (+ 2 channels + analytics)
Choose this if a shared inbox or ticket queue is drowning in repetitive items — and a human still needs to approve every reply, but shouldn't have to write each one from scratch.
| What you get | Classification + routing + draft-reply pipeline over one channel (shared inbox or IT ticket queue). Human-in-the-loop, nothing auto-sent. |
| Timeline | 2 weeks |
| Success measure | On 200 historical items, ≥80% correct classification/routing; ≥1 usable draft per category. |
| What it leads to | A full contact-center assistant across more channels, with analytics and ongoing automation operations. |
In 2 weeks we'll deploy a production-grade local-AI stack on your GPU — monitored, documented, ready to extend.
For: IT directors / CTOs who want to see a real self-hosted stack, not slides.
€7,000 Good, single GPU · €10,000 Better, multi-GPU + model catalog
Choose this if you want the platform, not the application — i.e. your IT team wants a hardened, monitored self-hosted LLM stack they own and extend, and you already have (or are buying) a GPU. Not this if what you actually want is answers over your documents — that's Pilot 1.
| What you get | The UbuntuLLM-derived stack on your hardware: vLLM/Ollama/llama.cpp (NVIDIA, AMD, Intel, or Apple Silicon/Mac), Open WebUI, Traefik TLS, monitoring, backups, runbook. 1–2 models served. |
| Timeline | 2 weeks (requires client hardware — GPU or Mac — available day 1) |
| Success measure | Endpoint live on your network; monitoring + backup verified; runbook handed over. |
| What it leads to | A production multi-model platform under SLA, with any automation (RAG, document AI, agents) built on top of it. |