01
Sources are stale or contradictory
The assistant may find several versions of a procedure. Without owners, dates and publishing rules, it cannot know which answer applies.
AI / company knowledge and RAG
Company AI knowledge assistants with source-grounded answers
We build company AI assistants and RAG systems with governed sources, citations, roles, evaluations, feedback, monitoring and cloud, hybrid or local options.
01answers grounded in retrieved sources with a safe path to say “I don't know”
02access roles that separate knowledge for different user groups
03an evaluation question set, feedback and quality monitoring after launch
AI development servicesBusiness context
A useful knowledge assistant begins with sources, access and evaluation — not a chat window.
RAG retrieves the parts of company knowledge needed for an answer and provides them to the model as context. It does not automatically mean the model is trained on the documents or that every answer is correct.
The design includes knowledge updates, user roles, answer sources, citations, an evaluation set and behaviour when material is incomplete or contradictory. Only then do we select the interface and model.
Where value leaks
Problems that cannot be solved by connecting a folder of documents
The first useful scope comes from the real bottleneck, not from a prebuilt package.
02
Everyone can see everything
One shared index without permission filtering can expose information to people who should not have access.
03
The answer sounds right but has no source
Without a citation or link, the user cannot quickly verify the procedure, figure or surrounding context.
04
There is no evaluation set
Testing a few convenient questions does not demonstrate quality. The set must include easy, difficult, incomplete and deliberately unanswerable cases.
Implementation scope
A knowledge, access and quality layer ready for daily work
The first pilot can serve one user group and a limited source set. Source quality and answer criteria matter more than document count.
Knowledge and question inventory
We identify sources, owners, users, recurring questions and cases that still require an expert.
Source import and preparation
We organise formats, metadata, versions and content segmentation required for effective retrieval.
Retrieval and answer generation
We design retrieval, context and answer instructions with a refusal threshold when evidence is insufficient.
Citations and source path
An answer can identify the document, section or link so the user can inspect the original material.
Roles and permissions
We design login and source filtering within the agreed scope, then test access for each role.
Evaluation, feedback and monitoring
We create a question set, scoring method, feedback capture and knowledge-update process for after launch.
Delivery
A RAG pilot built around real questions from the team
Each stage has a decision, an output and a clear reason to move forward.
01
Questions and criteria
We collect user groups, common questions, high-risk areas and conditions for a good or unacceptable answer.
02
Sources and access
We prepare a selected document set, owners, versions and visibility rules.
03
Pilot and evaluation
We test retrieval and answers against the evaluation set, including questions without a valid source.
04
Staged release
We verify permissions, collect feedback and expand knowledge only after quality is reviewed.
Expected outcomes
What a knowledge assistant should give the team
one place to ask questions about a selected knowledge domain
a shorter path to the correct procedure or document
an answer with a visible source or an explicit refusal
access limited to knowledge appropriate for the user's role
measurable quality across a stable question set
an update process instead of a one-off document import
RAG pricing / net prices
The first pilot begins with one knowledge domain and a measurable question set.
Cost depends on source count, update method, permissions, citation quality, login integrations and whether the architecture is cloud, hybrid or local.
| Option | Price | Scope | Included |
|---|---|---|---|
| AI opportunity audit | from PLN 900 net / one-off | The typical range is PLN 900–2,500. We review sources, users, questions, access rules and whether RAG is the right solution. |
|
| RAG knowledge assistant | from PLN 8,000 net / implementation | The typical range is PLN 8,000–25,000 for an assistant grounded in agreed sources with answer evaluation and quality controls. |
|
| Care and knowledge updates | from PLN 500 net / month | The existing RAG care threshold. Scope depends on source-change frequency, user count and the required monitoring level. |
|
| Advanced RAG | Custom quote agreed individually | Local or hybrid deployment, multiple repositories, complex roles, SSO integration, high scale or knowledge requiring enhanced controls. |
|
Net prices in PLN. “From” means the smallest sensible first-stage scope. Advanced delivery — including multiple integrations, migrations, custom roles and permissions, local or hybrid environments, high volumes, extended SLAs or business-critical workflows — is quoted individually after discovery or an audit.
Evidence
We state demonstrator status instead of inventing a client RAG case
The current portfolio confirms AI features working with user materials, not a full organisational RAG implementation with citations, roles and evaluation.
Adjacent evidence / not full RAG
Med2Be — an AI module working with notes and PDFs
The product confirms learning-material generation from user notes and PDFs. We do not present it as company RAG with role-aware retrieval and citations.
View the documented moduleCase study placeholder
Reserved for a verified RAG pilot
The future case will document sources, roles, the evaluation set, scoring method, access tests and limitations confirmed during delivery.
View the current portfolioDecision support
Related decisions about data, architecture and workflow
AI development services
The parent offer for audits, chatbots, assistants, documents, agents and automation.
Open resourceBusiness process automation
A related scope when knowledge work also needs rules, statuses and system actions.
Open resourceAI document automation
The solution for extraction, classification, validation and document data flows.
Open resourceFAQ
Questions before the first scope
Clear answers before technology, timing and budget are committed.
Is the model trained on our documents?
Usually not. In RAG, relevant materials are retrieved at answer time and passed to the model as context. Fine-tuning would be a separate decision and scope.
Can the assistant hallucinate?
That risk cannot honestly be eliminated. We reduce it with grounded sources, thresholds, instructions, evaluations, citations and a refusal path.
Can the assistant respect access permissions?
Roles and source filtering can be designed, but they must be integrated and tested for the actual repositories, accounts and access scenarios.
Can RAG run locally?
Sometimes. The choice between local, hybrid and cloud architecture depends on the data, infrastructure, organisational requirements, expected quality and operating cost.
How many documents can be connected?
Document count alone is not a useful criterion. Formats, quality, versioning, permissions, update frequency and the target questions matter more.
How much does a company RAG assistant cost?
An AI opportunity audit starts from PLN 900 net. A RAG knowledge assistant starts from PLN 8,000, with a typical implementation range of PLN 8,000–25,000. Care and source updates start from PLN 500 per month. Local, hybrid, multi-repository and complex permission models are quoted individually.
How do you evaluate answer quality?
Against an agreed question set with expected sources and criteria. We score retrieval, answer faithfulness, citation and safe refusal separately.
Next step
Describe the knowledge sources, users and recurring questions
Share the document types, user groups, access rules and questions that currently return to experts. We will assess whether RAG fits and how small the first pilot can be.
Related solutions
Connect visibility, service and operations.
E-commerce / sales and order operations
E-commerce stores designed around sales and order operations
We design e-commerce stores with catalogue UX, cart, checkout, payments, delivery, order operations, technical SEO and verified integrations.
ExploreSoftware / APIs and data flow
System and API integrations for a connected data flow
We connect business systems through APIs, webhooks and controlled data flows across CRM, ERP, commerce, panels, imports, logs, retries and monitoring.
ExploreAI / tasks, tools and control
AI agents for business with approvals, logs and human control
We design AI agents for defined tasks with scoped data, tools, rules, approvals, logs, exception handling and a safe human fallback.
Explore