LLM & custom AI
Ask your company's documentsquestions directly.
For teams that often search through procedures, contracts and manuals: we connect large language models (LLMs) to the company's documents and databases, so the answer comes quickly, together with the document it comes from, and only from the sources each colleague has access to.
Your company's knowledge already exists. It just needs to be easy to find.
In most businesses, the answers are already written down somewhere: in a procedure, a contract, an old email or a file on the drive. The information exists; what is missing is a short path to it.
Exact-keyword search doesn't work
If you don't know the document's title or the term its author used, the classic search in the drive or intranet doesn't help.
The same questions go to the same people
Experienced colleagues become the company's “knowledge base” and answer questions about procedures, prices or contract terms every day.
Slow onboarding for new colleagues
A new employee has to work through dozens of documents without knowing which are current. Weeks of questions pass before they find their way.
Long documents, read by hand
Contracts, tender specifications or reports have to be read in full to extract a few details. The work is repetitive and easy to get wrong.
Generic AI tools, without context
A public AI chat doesn't know your company's rules and can give a convincing but wrong answer. On top of that, it is unclear where the data you enter ends up.
What LLM integration is and how a document assistant works
LLM integration means connecting a large language model to the company's documents, data and applications, so that it can answer questions, search or extract information based on them. The focus is on knowledge: the assistant finds and explains information, and when the AI also needs to carry out actions in other applications, we are talking about AI agents. We use an approach called RAG (retrieval-augmented generation). Put simply: before answering, the system searches the data sources you have configured, picks the relevant passages and writes the answer based on them, showing where they come from. This can increase relevance and makes answers easier to check, but it does not eliminate all errors. That is why we also build the mechanisms that make them visible.
Selected data sources
Together we choose which documents and databases go into the system: procedures, contracts, product sheets, resolved tickets. Anything not configured is not consulted.
Semantic search
Texts are indexed by meaning (through embeddings), not just by words. The question “how long is the warranty?” also finds the paragraph about the “coverage period”.
Answers with source references
Each answer can include the document and passage used, so the user can quickly check the information before relying on it.
Role-based access
The system respects who is allowed to see what. A colleague in sales does not get answers from HR documents if they don't have access to them.
Continuous evaluation of answers
We test the system on a set of real questions with known answers and track user feedback, to spot and correct weak answers.
Dedicated LLM solutions
From integrating a model into your application to evaluating its answers: the building blocks of an AI system you can trust.
LLM integration
Language-model features added to the applications and processes you already use, with the model chosen for cost and confidentiality.
DetailsCustom AI assistants
Assistants built for a specific role — quoting, in-house legal, technical support — with that role's tone, rules and sources.
DetailsRAG systems
Assistants that search the configured sources before answering and show where each piece of information comes from.
DetailsEnterprise knowledge bases
The infrastructure that keeps company knowledge in order: inventoried sources, permissions, updates and an owner for each domain.
DetailsSemantic search
Search by meaning across your website, catalogue or documents: relevant results even when users do not type the exact words.
DetailsIntelligent document processing
The engine that extracts, classifies and summarises documents in batches and delivers structured data, ready to review and import.
DetailsAI API integration
Connecting AI providers' APIs to your systems, with keys kept on the server, limits, logs, cost control and a fallback provider.
DetailsAI evaluation and monitoring
Test sets, checks at every change, human spot-check reviews and cost tracking, so you know how your AI performs over time.
Details
What can change in the way you work
- 01
Answers found faster
Colleagues can ask the question in their own words and get a synthesised answer, instead of opening ten documents.
- 02
Experts are interrupted less
Routine questions can be answered from the documentation, and specialists stay available for the genuinely complicated cases.
- 03
Answers that are easier to check
The source reference lets the user confirm the information and notice when a document is out of date.
- 04
Structured data from unstructured documents
Information from contracts, quotes or reports can be extracted into clear fields, ready to be checked and imported into other systems.
- 05
The documentation gets better
Unanswered questions show what is missing from internal documents, so you know exactly what is worth writing or updating.
Results depend on the quality, currency and structure of the documents, as well as on how the system is used. We do not guarantee a particular level of accuracy; we measure it and improve it along the way.
Example use cases
The scenarios are conceptual examples showing what can be built. They are not delivered projects or results achieved for clients.
- Industrial manufacturing
An assistant for technical procedures
Technicians ask about maintenance steps or a piece of equipment's specifications and get the answer extracted from manuals and work instructions, with the source page shown.
Conceptual example - Legal or tax consultancy
Semantic search in the document archive
Consultants find clauses, opinions or similar cases by meaning, not by title. The system suggests relevant passages, and the interpretation stays with the specialist.
Conceptual example - Construction and tenders
Extraction from tender specifications
The requirements, deadlines and qualification criteria are extracted from a tender's documentation into a table, which the team checks before preparing the bid.
Conceptual example - Retail with a store network
A knowledge base for staff
Store employees get answers to questions about the returns policy, promotions or internal procedures, from documents approved by head office.
Conceptual example
What the project can include
We build anything from a simple assistant on a set of documents to custom AI solutions integrated into your applications.
Assistants and search
- LLM-based assistants for internal teams or customers
- RAG systems on the company's documents and data
- Chatbots connected to the company's documents
- Semantic search across archives and drives
Knowledge and documents
- Smart knowledge bases, with role-based access
- Extracting and structuring information from documents
- Automatic document classification
- Summaries of long documents, with source references
Integration and quality
- Integrating AI models via API into existing applications
- Evaluating and optimising AI answers
- Test sets and quality monitoring over time
- Automatic updating of indexed sources
Custom AI solutions
- AI workflows built around a specific company process
- Choosing the right model for cost, speed and quality
- Dedicated interfaces or integration into existing tools
Indicative prices
Packages and prices
Price benchmarks for this service, in euro, excluding VAT. The final offer is set after a conversation about the project.
See all pricesAI assistant on your website
+ from €90/month operation · 2–4 weeks
from€1,500
- Answers from your documents and pages
- Takes requests and passes them to the team
- Installed on the website, WhatsApp or Messenger
- Monthly monitoring and improvement
- Recommended
Automation
n8n or Make · 1–3 weeks
from€900per workflow
- Analysis of the process and the data
- Connecting the applications you already use
- Parallel testing before going live
- Documentation and training
- Optional operation, from €150/month
AI voice agent
+ per-minute costs · 4–6 weeks
from€2,900
- Handles calls, bookings and frequent questions
- Transfers to a person when the situation requires it
- Integration with your calendar or CRM
- Disclosure and consent in line with the law
AI provider costs (OpenAI, Anthropic, telephony) are billed on consumption, transparently. The online AI assessment is free.
How we implement it
Step 01
Inventory of sources
We establish which documents and data go into the system, who has access to them, how often they change and which of them are genuinely current.
Step 02
A set of real questions
We collect questions the team already asks, together with the correct answers. They become the benchmark we evaluate the system against.
Step 03
Preparing and indexing the data
We clean, split and index the documents, keeping the important metadata: author, date, department, access level.
Step 04
Building the assistant
We configure the model, the instructions, the rules for unanswerable questions and the interface, whether that is an internal chat, a Slack integration or a dedicated application.
Step 05
Evaluation and adjustment
We run the test set, analyse wrong or incomplete answers and adjust the search, sources or instructions until they reach a level you accept.
Step 06
Launch and maintenance
We start with a pilot group, collect feedback and establish how the sources are updated when documents change.
Compatible data sources and models
The assistant can read from the places where you already keep your documents and can work with different AI model providers, depending on your cost, quality and confidentiality requirements.
The compatibility of each source, the document formats and the hosting options are checked for each project.
- Google Drive
- Microsoft SharePoint
- Microsoft 365
- Notion
- Confluence
- Slack
- SQL databases
- PostgreSQL
- PDF and Word files
- OpenAI API
- Anthropic API
- Privately hosted open-source models
Security & control
Access, sources and answers under control
An assistant connected to company documents must follow the same access rules as the rest of your systems. We design it with that requirement as the starting point.
What data reaches the AI provider
Only the passages relevant to the question are sent to the model, not the whole archive. We choose providers and settings that do not use the data for training, according to their terms, and for sensitive data we also consider privately hosted options. The legal basis for processing personal data is established in line with the GDPR.
Role-based access control
Permissions from the original sources are respected or explicitly reconfigured, so that each user gets answers only from the documents they are entitled to.
Up-to-date sources
We establish how and how often documents are re-indexed, and old versions can be withdrawn so the assistant doesn't answer from superseded procedures.
Questions without enough information
When the sources don't contain the answer, the assistant is instructed to say so clearly and tell the user who they can ask, instead of filling in the gaps.
Logging and evaluation
Questions, the sources used and feedback can be recorded, in line with internal rules, for auditing and for improving the quality of answers.
Useful guides
Before you decide, read
- LLMs & dataWhat is RAG? A simple explanation for AI assistants on company documentsRAG means the AI assistant first searches your documents and only then answers, citing the source. What it solves, what it doesn't and how to evaluate it.Read the guide
- LLMs & dataThe OpenAI API for businesses: what it is and how to bring ChatGPT into your processesThe OpenAI API is how a company's applications use the models behind ChatGPT. How it differs from the ChatGPT app, typical integrations, how data is handled and how the cost is calculated.Read the profile
- LLMs & dataClaude AI for businesses: what it is, what it's for and how it compares with ChatGPTClaude AI is the family of language models developed by Anthropic. What it can do for a business, the difference between the app and the API, how data is handled and the limits you need to know.Read the profile
- ComplianceUsing ChatGPT at work without the risks: an AI use policy for employeesHow to use ChatGPT and other AI tools in your business without exposing confidential data: what you never enter, which plan to choose, what the GDPR requires and what a simple internal policy looks like.Read the guide
Frequently asked questions
What is a RAG system, in short?
It is an AI assistant that, before answering, searches for the information in the documents and data you have configured and builds its answer from them. This makes the answers more relevant to your company and lets them be checked against the source.
What is the difference between a document assistant and an AI agent?
A document assistant (an LLM with RAG) answers questions, searches and extracts information from the configured sources, with a reference to the document. An AI agent also uses authorised tools to carry out actions, for example creating a task in the CRM. Often, the document assistant is the foundation on which an agent is built later.
If we use RAG, can mistakes still happen?
Yes. RAG reduces the risk of made-up answers but does not eliminate it: the system may pick an unsuitable passage or misinterpret a document. That is why we show the sources, test on real questions and recommend human review for important decisions.
What happens if the AI doesn't know the answer?
The assistant is configured to recognise situations where the documents don't contain the information and to say so directly, possibly pointing to the right person or department. These questions are recorded and show what is missing from the documentation.
How is the data protected?
We work with role-based access, send the model only the passages it needs and choose providers and settings suited to the level of confidentiality. For sensitive data we can consider privately hosted models. GDPR compliance is assessed together with you for each project.
What happens when the documents change?
We set up an update mechanism: periodic re-indexing or re-indexing on every change, plus the option to withdraw old documents. The quality of the answers depends directly on how current the sources are.
How do I measure the assistant's effectiveness?
We use a set of test questions with known answers, user feedback and usage indicators: how often an answer is rated useful, how many questions remain unanswered and which sources are used most often.
Which AI model do you use?
We choose the model based on your requirements: answer quality in Romanian, cost, speed and confidentiality. We can work with APIs from providers such as OpenAI or Anthropic, or with separately hosted open-source models.
What does the cost of a project like this depend on?
On the volume and format of the documents, the number of sources and integrations, the access and security requirements, the chosen model and the estimated volume of questions. After an initial analysis we can present the options and their associated costs.
Can’t find the answer? Write to us and we’ll reply directly.
Make your company's documents easy to query.
Tell us what types of documents your team uses and which questions come up most often. We will propose a suitable approach, including how we check the quality of the answers.
Related services
- AI agentsAI assistants that follow the steps you set, use only approved tools and complete concrete tasks, with a person making the decisions.
- AI automationAutomated workflows between the apps you already use, with AI where it helps and human review where it matters.
- Custom software & integrationsApplications, platforms and integrations built around your processes, so data flows between systems on its own.