Skip to content

From AI ambition to AI in production.

KARPINITY is an AI innovation lab in Zurich. We work with your team to find where AI can grow revenue or cut costs, then take the idea all the way to a product your people use every day, with the business case checked at every stage.

Start a projectSee the six stages

KARPINITY · Live

Simulated

Agent
Invoice agent
Autonomy
Autonomy · rung 3 · Act with approval
Owner
Accounts payable lead

Warehouses · Switzerland

  • Warehouse · Zurich · Live
  • Warehouse · Basel · Live
  • Warehouse · Bern · Live
  • Warehouse · Lausanne · Live
  • Warehouse · St. Gallen · Pilot
  • Warehouse · Lugano · Pilot
  • Warehouse · Geneva · Planned
  • Live
  • Pilot
  • Planned
Invoices today
1,284
Prepared for approval
1,231
Handed to a person
53
Evaluation set
Passing

Cost per task · within budget

Run feed

  1. 09:41:07 · Invoice received · Basel · 3 documents read
  2. 09:41:09 · Matched to order and goods receipt
  3. 09:41:10 · Coded · prepared for approval
  4. 09:41:22 · Invoice received · Zurich · 3 documents read
  5. 09:41:24 · Mismatch · invoice 40 units, goods receipt 36
  6. 09:41:25 · Guardrail · IBAN differs from supplier record
  7. 09:41:25 · Handed to the accounts payable lead · handoff package attached
  8. 09:42:03 · Batch approved by a person · logged

  1. 65%of Swiss companies say AI is part of their long-term strategy.

    Source: CorpIn with the University of St. Gallen · Swiss AI Report · March 2025 · 1,338 managers

  2. 13%run their AI projects against clearly defined, measurable KPIs.

    Source: same report

  3. 14%scale AI systematically across several areas of the business.

    Source: EY Switzerland · AI Survey 2026 · May 2026 · 604 companies

The distance between the first number and the last is where we work.

The number in between explains the gap: without a measurable KPI, a pilot cannot prove its value, so it never earns the budget to scale. We anchor every project to a measurable KPI and a clear business case from day one, so what we build is made to scale into production.

A lab with two engines.

For clients

We help you find where AI has real potential in your business: use cases that are feasible, can scale and deliver business value.

We take the most promising through six stages, from the business case to a system your team runs. Each stage ends with a decision you make with the evidence in front of you.

How we work
  • NauvoIn the App Store
  • TelebimmerLive
  • ZezaBeta
  • KutioBeta
  • ParkberryBeta

Proven in our own products

Every method we apply is proven over years of designing, building and shipping products of our own.

When we recommend a solution, an innovative approach or an on-device AI model, we have already built it and proven it in our own products.

See the products

Watch one idea cross.

Illustrative project · supplier invoice exceptions at a logistics company

  1. 01Ground the case3-5 weeks
  2. 02Prove the value3-6 weeks
  3. 03Engineer the system8-16 weeks
  4. 04Pilot in one place6-12 weeks
  5. 05Scale what works3-6 months
  6. 06Run and evolveOngoing
InvoicePurchase orderGoods receiptInvoice agentHuman approvalPaymentInvoicePurchase orderGoods receiptInvoice agentHuman approvalPaymentMatchFlagDraft queryGuardrailAGT·INV·01Autonomy · rung 3Pilot footprintData protectionSecurityHuman oversightMonitoringLive · one warehousePaymentInvoice agentCaseExpectedResultQuantity mismatchFlagIBAN changedHand overAmount above CHF 10,000Hand overBaselineKPI fixed in stage 01cost of one exceptiondata: usable01Site 02Site 03Site 04Site 05Site 06Site 07Q1Q2Q3Q4Quarterly reviewIn productionInvoicePurchase orderGoods receiptInvoice agentHuman approvalPaymentInvoicePurchase orderGoods receiptInvoice agentHuman approvalPaymentMatchFlagDraft queryGuardrailAGT·INV·01Autonomy · rung 3Pilot footprintData protectionSecurityHuman oversightMonitoringLive · one warehousePaymentInvoice agentCaseExpectedResultQuantity mismatchFlagIBAN changedHand overAmount above CHF 10,000Hand overBaselineKPI fixed in stage 01cost of one exceptiondata: usable01Site 02Site 03Site 04Site 05Site 06Site 07Q1Q2Q3Q4Quarterly reviewIn production

01 · Ground the caseA sketch, a business case and a verdict on the data. You decide whether the idea is worth proving.

02 · Prove the valueWe write the evaluation first. Then we build the prototype that has to pass it.

03 · Engineer the systemProduction engineering for the pilot’s footprint: security, data protection, human oversight, monitoring.

04 · Pilot in one placeReal users at one site, measured against the KPI fixed in stage 01.

05 · Scale what worksMore sites and teams, the engineering wider use needs, and the value measured again.

06 · Run and evolveWe run it with you, review the value every quarter and feed what we learn into the next case.

One idea, six decisions, one system in use.

Agents you can hand real work to.

Every agent we build gets a persona, the way a new hire gets a job description: a role, the tools it may use, the decisions it may take, and the moment it hands over to a person. Your accountable executive decides how much it may do on its own, and it has to pass an evaluation before anyone relies on it.

How we engineer agents

Simulated run

Human HUM·APL·01

Accounts payable lead

  • Approves payments and exceptions
  • Owns supplier relationships

Agent AGT·INV·01

Invoice agent

Autonomy · rung 3 · Act with approval

Rung 3 · it acts after a person approves. The accountable executive chooses the rung.

Role
Reads, matches and prepares supplier invoices for payment
Reads
invoices, purchase orders, goods receipts
May
match documents, code invoices, flag mismatches, draft the supplier query
May not
release a payment, change supplier bank details
Hands over when
the amount is above CHF 10,000, or the supplier’s IBAN has changed
Proves
every step logged with the documents it used
  1. 14 invoices received · 42 documents read
  2. 12 matched to order and goods receipt · prepared for approval
  3. Mismatch found · quantity 40, goods receipt 36 · supplier query drafted
  4. Guardrail · IBAN differs from supplier record · stopped
  5. Handed to the accounts payable lead · handoff package attached
  6. Batch approved by a person · logged

One partner, from the idea to the running system.

We shape the idea with you, then carry it through prototype, scale and operation.

  1. Stages: 01, 02, 03, 04, 05, 06

    New AI-native digital products

    New AI-native digital products and revenue streams, from the first idea to daily operation.

    Diagram: one idea travels the six stages of the KARPINITY Method, with one partner across the whole arc. 01 Ground the case, 3-5 weeks, turns it into a business case. 02 Prove the value, 3-6 weeks, into a prototype measured against its evaluation. 03 Engineer the system, 8-16 weeks, into a governed system for one site. 04 Pilot in one place, 6-12 weeks, puts it in front of real users. 05 Scale what works, 3-6 months, extends it across a wider scope. 06 Run and evolve, ongoing, runs it with a live dashboard. A decision gate, continue, adjust or stop, sits between each pair of stages, and a return line brings the next idea back to the start.

  2. Stages: 01, 02, 03, 04, 05, 06

    Agentic platforms

    AI agents on web, mobile, messaging and the back office, built, deployed and run on one platform.

    Diagram: one platform serves four channels: web, mobile, messaging and the back office. Inside it, agents with persona codes, each with its rung on the autonomy ladder and a stop control, work with tools, knowledge and logs. Every action leaving the platform passes the guardrails. Payments and bank details go to a human approval queue; reversible actions go out with every step logged. Underneath, monitoring tracks cost per completed task, handover rate and drift.

  3. Stages: 01, 02, 03, 04, 05, 06

    Document intelligence

    Enterprise document processing and data extraction, checked against your systems, with a person on every uncertain case.

    Diagram: documents arrive as PDFs, scans and email: an invoice, a contract and a form. The system reads their text, layout and tables, then extracts each field, such as the supplier, the IBAN and the total, together with the quote and page it came from. Every field is checked against the ERP and master data: here the order and the goods receipt match, but the IBAN does not. Confident documents are posted; uncertain ones go to a review queue where a person decides, and every correction becomes a new evaluation case.

  4. Stages: 01, 02, 03, 04, 05, 06

    AI customer experience

    Concierge agents, personalisation and voice on every channel your customers use, with your staff one step away.

    Diagram: a customer journey across five touchpoints: discover on the website, ask on messaging, book in the app or by QR code, use by voice, and a follow-up message. At each touchpoint an agent answers (concierge, recommender or voice), and all agents draw on one shared knowledge layer. At two touchpoints an agent hands the case to staff: the concierge passes on an action to take, the recommender an insight about the customer. A dashboard shows cost per channel and the most frequent customer questions.

  5. Stages: 01, 02, 03, 04, 05, 06

    Edge AI & connected intelligence

    Edge AI and connected intelligence: sensors, vehicles and smart devices joined to AI on the device, at the edge or in the platform.

    Diagram: five kinds of source on the left: sensor data such as heat and vibration, IoT systems on machines and fleets, an ERP with orders and stock, AI agents with their actions and logs, and external data such as weather and markets. Their data goes to models at the edge, with a local cache, that keep working offline, inside a boundary where the data stays on your network. A platform gives every reading its source and age, applies alert rules and answers questions in plain language. On the right, people act on it: an alert in a messaging app, a map, a live display, and an operator who decides.

  6. Stages: 01, 02, 03, 04, 05, 06

    AI guardrails

    Governance, compliance and security for AI in production, with evidence an auditor can read.

    Diagram: a system you already run, under test. A case set, each case with its task, expected outcome, grader, class and source, runs several times: one case fails on its third run. Code, model and person graders score the outcomes. Red-team findings cover prompt injection, tool misuse and data leakage. An evidence pack with seven tabs collects the results. A release gate lets the system into live use, where every change reruns the set and live failures become new cases.

How every project is built: five foundations, from agent design to compliance

Made in the lab.

Eighteen products we design, build and run ourselves.

All products

Projects delivered to our clients.

Since 2015, innovation projects for our customers: proofs of concept, investor demos and production systems.

revFADP · EU AI Act · GxP

Read against your use case in stage 01, before anything is built.

AI regulation in Switzerland
All work

KARPINITY · innovation infinity

What do you want in production?

Innovation for our clients, and for our own products.

Your sentence goes with you to the contact form. Nothing is sent until you submit it there.

Zurich · Yerevan · Nis · Skopje

One line · the KARPINITY film

The idea behind the film · 37 seconds, no sound

The film follows one line from start to finish. It begins as a pencil stroke in front of the Matterhorn, because every project with us starts as a sketch. The line then traces the Landwasser Viaduct on the Rhaetian Railway’s Albula line, a stone viaduct built more than a hundred years ago that trains still cross every day. We hold the systems we build to that standard.

Halfway across, the wireframe train turns into the real red train: the point where a prototype becomes a system in production. On the phone in the carriage, the status moves from Prototype to Pilot to Live, the stages a client’s project goes through with us. At night the line leaves Zurich, where the lab is based, spreads across Switzerland as a network and closes into the KARPINITY mark, which is itself drawn as one woven line.

What the film shows: the Matterhorn at first light. A pencil line runs along a railway track and sketches a stone viaduct, which fades into the real Landwasser Viaduct at blue hour. A train is drawn on it as a wireframe, its windows light up, and the drawing becomes the real red train crossing. Inside a carriage, a phone shows a sketch of an app that becomes a finished interface: a map of Switzerland, with a status that reads Prototype, then Pilot, then Live. The map fills the frame at night, a network of lines draws out from Zurich, and the network draws in to become the KARPINITY mark. End card: KARPINITY · innovation infinity.