Data and context · AI Engine
Models change. The architecture stays yours.
AI Engine is your company’s AI architecture. We design and build it process by process: data, procedures, decision rules and system connections in one place, with access control. People with an AI assistant and agents, ours and other vendors’, work on it.
An architect’s view
Most companies bolt AI on. We start with the foundation.
An assistant in email, an agent in the helpdesk, a chatbot on the website. Each tool stands on its own, and each gets its own piece of company knowledge from its vendor.
With every new tool, the same discount rules live in several places and nobody knows which version is current. Every new agent starts from zero. When an experienced employee leaves, their way of working leaves with them.
That is why we design the other way round. First a shared foundation: data, procedures, decision rules and system connections. On top of it we put assistants and agents, ours and other vendors’. Each next one uses what is already in place.
How it works
Systems feed AI Engine. Every correction flows back into it.
We do not replace your systems. ERP, CRM, Microsoft 365, the helpdesk and catalogues stay, and AI Engine organises from them what people and agents need.
The systems you already have
Data on clients, products, orders and documents comes from your systems. For each source we name the owner, the data scope and the permissions.
Company context with permissions
In AI Engine, data meets procedures and decision rules. Everyone sees only what they have access to, and every step is logged.
The learning loop
Approvals and corrections from daily work flow back into AI Engine once the process owner signs them off. This updates the context. It does not train a model on your data.
Systems you already have
- ERPproducts, orders, inventory
- CRMcustomers, quotes, contacts
- Microsoft 365Outlook, SharePoint, Teams
- Helpdesktickets and service history
- PIM & cataloguesparameters, prices, availability
Company context
AI Engine
- Context Lakehousedata about customers, products and orders
- Skillshow work gets done at your company, written down step by step
- Decisionscriteria and past decisions
- IntelligenceAI models
With permissions and a trail: everyone sees only what they have access to, and every step is recorded
Who uses it
- Boardsteers the company more easily
- Managersfaster decisions
- Whole teamaccess to knowledge, faster work
- Agentswoven into processes
- People’s approvals and corrections flow back into the context
Architecture
Six elements. You replace the model, the rest stays yours.
We design AI Engine like a building: from the foundation up. Five elements are the structure that stays in the company for years. The AI model is replaceable, because models change faster than your processes.
Human+AI Teams
Who does the work?
people and agentsPeople with an AI assistant and agents, ours and other vendors’. Each has a defined scope of work and a person responsible for the result.
From a project · Pharmaceutical company, AI adoption30 AI ambassadors from more than 10 departments.See the case studyDecisions
Why this way and not another?
stays with youCriteria, exceptions and earlier rulings with their reasons. With them, AI supports decisions in line with company rules, and every decision can be explained.
From a project · Financial company, Document processing · In progressEvery analyst decision (approve, correct, reject) stays in the change history.See the case studySkills
How do we do it?
stays with youWays of doing tasks, written down step by step. The process owner approves every change, and every version is recorded.
From a project · Manufacturing company, QuotingWhat a few experienced consultants knew about turning an item into a product is written down: repeatable items are recognised by rules, without the model.See the case studyIntelligence
Who reads, writes and analyses?
replaceableAI models chosen for the task and the sensitivity of the data, in the cloud or on your server. Changing the model does not require rebuilding the rest.
From a project · Financial company, Private AI environment · In progressLocal open-weight models, including the Polish Bielik, on a GPU server on the internal network.See the case studyMCP
How does AI reach your systems?
stays with youControlled connections to ERP, CRM, email and documents. For each system we describe the allowed reads and writes, and the integration system enforces permissions, not the wording of a prompt.
From a project · Manufacturing company, Company knowledgeTeams, SharePoint and the PIM connected through n8n: the gateway decides where the answer comes from.See the case studyContext Lakehouse
What do we work on?
foundationCompany data and knowledge in the right scope: clients, products, orders, documents. Each source has an owner, and documents keep their origin and version.
From a project · Manufacturing company, Product selectionpilot target: product selection timeSee the case study
Design principles
The model reads and understands. Rules and your systems do the maths.
We follow these principles in every implementation on AI Engine.
Rules first, then the model
Whatever can be written as a rule is handled by a rule: cheaper, faster and without the risk of a mistake. The model only gets what the rules do not cover.
The system has the final say
Prices, totals and deadlines are calculated by the ERP and by code, not by the model. An agent’s output is checked before it moves on.
Knowledge in data, not in code
A new product, procedure or rule is a new record, not a new implementation. The logic lives in data that can be reviewed and corrected.
The system enforces permissions
An agent sees and does only what the integration system allows, whatever the prompt says. Every step is logged.
People decide at agreed points
Before the start we agree where a person decides: important results, exceptions and changes to rules. The agent works within an agreed scope.
The model fits the task
Each step can use a different model: a stronger one for hard analysis, a cheaper one for simple tasks, a local one for sensitive data. Replacing the model does not change the rest.
How we build
One process to start. The next ones on a ready foundation.
We do not start by putting the whole company in order. The first process delivers value and the first part of the foundation, and every next one builds on it.
A map of processes and decisions
We choose a process the company repeats every day, with an estimated value. Before the start we write down the baseline, the target and the success threshold.
Starting context
We organise only what this process needs: sources, data owners and permissions. We agree integrations and operating limits with IT.
Acceptance on examples
We check the result, the source, approvals, access scope and the ability to roll back a change. We measure time, cost and the share of tasks that need correction.
The next process on what exists
The next agent takes the data, connections, Skills and rules already in AI Engine and adds only its own part.
How AI Engine grows
Indicative. Scope and pace depend on the company’s readiness and the results of each stage.
01 · First 90 days
Starting context
Sources for the first task, data owners and access permissions.
02 · 3-12 months
A shared foundation
Context Lakehouse, consistent definitions and context from everyday work.
03 · 12-36 months
Knowledge that stays
The learning loop supports further tasks. New agents reuse existing context.
For CTOs and IT
We describe the architecture, integrations, permissions, activity history and acceptance criteria in a technical appendix. We agree them with IT before the first process goes live.
Appendix for CTOs and ITCase studies
Already running on AI Engine at our clients.
We describe clients anonymously. The status shows the stage of each implementation.
The quote reaches the client before the competition’s.
7 days → 1 day
response time for enquiries handled by the agent
See the case study
Financial statements and reports without manual retyping
80-90%*
about 3 min instead of 15-30 min of retyping
See the case study
Product selection in a few minutes, rather than an hour.
up to 1 h → <10 min
product selection time
See the case study
Company knowledge available in Teams
14 weeks
from kickoff to a Knowledge agent in Teams answering from FAQs, documents and product data
See the case study
Work in progress. These are results verified at the test stage or the targets we are aiming for. We will confirm them on completion.
Questions
Frequently asked questions.
How is AI Engine different from a data warehouse?
A warehouse stores data for reporting. AI Engine uses data but adds what AI needs to do the work: written procedures (Skills), decision rules (Decisions), controlled connections to systems (MCP) and models chosen for the task (Intelligence). Its foundation is the Context Lakehouse, which we build on Azure and Databricks.
Can other vendors’ agents use AI Engine?
Yes, within an agreed scope. Assistants and agents use AI Engine through MCP and connectors, with permissions and every step logged. The scope depends on each tool’s interfaces, so we confirm it with IT after checking the data, permissions and interfaces. More about system integrations.
Does the AI model learn from our data?
No. AI Engine grows by updating the context and the way of working. Corrections, approvals and decision rationales go through the process owner’s review, a check on examples and versioning. Only the approved version is used in further tasks. This is not automatic model training on company data.
What stays with us if we change the model or the partner?
Data, procedures, decision rules and system connections. The AI model is replaceable and the knowledge gathered in AI Engine stays in the company, so changing the model or the partner does not mean building from scratch.
Does our data need to be organised first?
You do not need to organise the whole company before the first use case. We start with the scope that is needed. Missing context is gathered through work, while process owners review and approve changes.
Where does AI Engine run?
Azure is the starting point. The most sensitive data can be processed by local models on company servers. We select other clouds and fully on-premise environments together with IT. More about the AI environment.
How long does it take to build AI Engine?
We do not build all of it before the first use case. The starting context for the first process takes roughly the first 90 days, and the shared foundation grows with each next process. Scope and pace depend on the company’s readiness and the results of each stage.
Let’s talk about the AI architecture of your company.
We start with one process and with what you already have: systems, data and AI tools. Together we will decide where to start building the foundation.
Book a callOr write to: hello@wearefuture.ai
