Deliverables
- Working assistant or processing pipeline
- Indexed, permission-aware knowledge base
- Test set of real questions and documents
- Admin and user guide
SERVICES / AI / AI assistants & document processing
AI assistants & document processing
AI assistant development for your business: assistants that answer from your company's own documents and show their sources, and processing that turns scanned or PDF documents into structured data. We work with the Anthropic Claude and OpenAI ChatGPT APIs, using the right model for each task.
assistants – zsh
$ scaffold build --assistant✓ indexed docs with access control✓ grounded answers with citations✓ extracted fields & flagged doubts→ assistant live, sources on every answer$
What you get
- An internal assistant that answers from your policies, manuals and documents
- Source citations on every answer, so staff can verify the information
- A customer support chat on your website that hands over to a person when needed
- Data extraction from invoices, contracts and forms, with uncertain cases flagged for review
- Permission handling: everyone only reaches the documents they already have access to
How it works
Sources & questions
We collect the documents and the typical questions or document types and agree on what counts as a correct answer.
Build & test
We test the assistant on real questions and documents and refine it until the answers are accurate and verifiable.
Launch & keep current
We roll it out to your team or customers and set up the knowledge base to update automatically with new documents.
Good fit when
Your staff spend a lot of time searching internal policies and old emails.
Your support team answers the same questions over and over every day.
Data from invoices, contracts or forms is still typed into a system by hand.
FAQ
What if the assistant gives a wrong answer?
Answers are grounded in your own documents and each one shows its source, so they are easy to check. If the documents do not contain the answer, the assistant says so instead of making something up.
Which languages does it handle?
Hungarian and English, and current models handle most European languages well. We test separately on Hungarian documents and questions, because quality has to be proven on real examples there too.
Where are our documents stored?
In hosting of your choice, in the EU where possible. Only the passages needed for an answer are sent to the AI service, and we choose provider settings under which your data is not used for model training.
Related services
All services in AI & automationAI adoption & strategy
Assessment, use cases and a rollout plan for Anthropic Claude and OpenAI ChatGPT, with data protection rules and team training.
Agent workflows & automation
AI agents that call tools, move data between systems and complete multi-step tasks, with approval checkpoints, guardrails and logging.
Code review & quality audit
A review of your codebase: architecture, security, performance, dependencies and tests, in a report ranked by severity with suggested fixes.
Technical documentation
Architecture overviews, API references, runbooks and developer onboarding for existing codebases, versioned alongside the code.
Test automation & QA
Unit, integration and end-to-end tests, accessibility checks and quality gates in your CI pipeline, for new or existing software.
Where could AI help your business?
Ask for a short assessment and we will show you where to start.