When the answer is somewhere in a large collection of documents, finding the right passage is only part of the task. People also need its context, the applicable version and a way to check what it says.
We develop retrieval-augmented generation (RAG) solutions that connect natural-language questions with relevant information and source-linked answers. Legal RAG is our strongest current specialisation, supported by hands-on experience in Hungarian legal information systems.
Legal materials have their own structure and terminology. A useful search solution needs to account for provisions, cross-references and validity over time. Our work focuses on helping users ask questions in everyday language and follow answers back to the relevant legal sources.
The same need for context and traceability arises in policies, procedures, technical manuals and internal guidance. We can build knowledge solutions around selected repositories, with access rules and update processes matched to the organisation.
RAG brings retrieval and language generation together. The system finds relevant material in a defined collection, then uses that material to help generate an answer. References let the user inspect the supporting sources.
Process documents into searchable content while retaining useful structure, source identifiers, dates and version information.
Use the question and its context to find material within the user’s permitted scope, including any applicable date or document filters.
Provide the retrieved material as context for the language model, with instructions on answer scope and handling insufficient evidence.
Present references alongside the answer so users can review the underlying passages and decide whether further investigation is needed.
Source references support verification; they do not guarantee that an answer is complete or correct.
Document quality and organisation affect what a RAG system can retrieve. We assess the available formats, structure, duplication and version history before choosing how to process and index the content.
Together, we define which sources belong in the solution, who owns them and how changes should be reflected. The design also needs a clear approach to document permissions, replacements and withdrawals, so outdated or inaccessible material does not become an unintended source.
Test representative questions against the source collection. Check whether retrieval finds the right passages and whether references support the answer, including cases where dates or document versions matter.
Define how the system responds when information is missing, conflicting or outside its scope. Decide where it should ask for clarification or direct the user to a person who can help.
Include permissions and restricted-content scenarios in evaluation. Agree which tasks require human review and how issues will be reported and investigated during operation.
A RAG solution can be delivered through a dedicated search interface or integrated into an existing application or assistant. We design the connections to document repositories, identity systems and business applications around the agreed scope.
Cloud and on-premises deployment are supported across mainstream platforms. Model choice and infrastructure depend on data boundaries, response-time expectations, operating cost and maintenance requirements. Support arrangements are tailored to the customer.
Explore our software engineering servicesiCode was the lead developer and solution architect for this AI search service, offered to the public by the Hungarian government. It answers natural-language questions about Hungarian legislation, including historical versions, with links to the source materials.
Built for large-scale public use, the system supports many simultaneous users and delivers responses within 15 seconds. It is one selected example from iCode’s wider software development history since 2003.
Read about the AI search projectTell us who needs answers, which information they rely on and what a useful result would look like. We can help define a focused starting point for assessment and validation.
If the task also involves taking action in other systems, explore how an intelligent assistant can combine knowledge access with a controlled workflow.
Discuss your RAG project Intelligent assistants and workflow automation