MedTech · Internal process automation
Hundreds of product documents, answered in seconds with the source
apoQlar’s support and operations teams spent a large part of their day searching hundreds of product documents by hand to answer customer and institutional inquiries. A generative AI assistant now takes the question in plain language and returns a precise answer in seconds, with a reference to the document it came from.
Key results
Client: apoQlar
A MedTech company in Hamburg developing mixed reality and AI for healthcare. Their platform brings holographic imaging into the operating theatre.
- Industry
- MedTech
- Use case
- Internal document retrieval and Q&A
- AI approach
- LLM with RAG and a vector database
- Users
- Support and operations teams
- Documents
- Product docs, reference materials
- Key feature
- Source-referenced responses
In short
Product knowledge that was locked inside the documents
- apoQlar’s products ship with hundreds of documents: technical specifications, regulatory materials and reference guides.
- Answering one customer or institutional inquiry meant searching them by hand, and the answer depended on how much time someone had and which documents they happened to check.
- A retrieval assistant now takes the question in plain language and answers from the entire document base in seconds, naming the document and section it drew from.
- The team verifies instead of searches, and because the source is attached the answer can go straight into customer-facing communication.
The starting point
The challenge
Critical product knowledge was locked inside hundreds of documents. Getting an answer required manual search, and the quality of that answer depended on how much time someone had and which documents they happened to check.
apoQlar develops mixed reality solutions for healthcare, and those products come with extensive technical documentation, product specifications, regulatory materials and reference guides. When the support team received a customer inquiry or an institutional request, finding the right information meant searching hundreds of documents by hand.
That was slow and it was inconsistent. Two people could come back with different information, or miss a relevant detail entirely, depending on which documents they opened.
The time spent searching was time not spent helping customers or preparing documentation.
The build
What we built
We built a generative AI assistant that sits on apoQlar’s internal document base and takes questions in plain language, the way someone would ask a colleague who had read every document.
A vector database instead of keyword search
Every internal document was processed, chunked into logical sections and embedded into a vector database. A question retrieves the most relevant passages with minimal latency, which keyword search matches on neither speed nor precision.
Every answer names its source
Each response carries references to the specific documents and sections it drew from. That was a deliberate decision: in MedTech the team has to verify and cite, not trust an AI-generated response. It also makes the output usable directly in customer-facing communication.
Context that survives a conversation
Long documents fill a context window quickly. The system keeps the conversation state while holding only the most relevant passages in the prompt, so accuracy and response time both hold up across several turns.
Text processing built for messy formats
Product documentation does not arrive in one clean format. We developed extraction, cleaning and grouping methods for the different document types, so retrieval works regardless of how the original was structured.
What changed
The results
Before
Manual search through hundreds of documents for every inquiry. Slow, inconsistent, and dependent on individual team members knowing where to look.
After
A question in plain language and an answer from the entire document base in seconds, with the source attached. Consistent, verifiable, and open to everyone on the team.
The support team moved from manual document search to conversational retrieval, and response times to customer and institutional inquiries improved significantly.
Because every answer carries its source, it can be trusted and cited. In a regulated MedTech environment that is the difference between an answer someone can send and an answer someone has to check first.
The assistant also surfaces connections between documents that manual search would have missed, which opens the accumulated product knowledge to everyone, not only to the people who have been there longest.
The GenAI-powered virtual assistant has been a game-changer for our internal support team. It revolutionized the way we access and utilize information from our vast document base.
Questions about this project
Internal document retrieval and Q&A. Critical product knowledge was locked inside hundreds of documents. Getting an answer required manual search, and the quality of that answer depended on how much time someone had and which documents they happened to check.
From question to answer across the whole document base: Seconds. Every answer names the document it came from: Sources. Vector retrieval instead of keyword search: RAG. Conversation state kept without losing accuracy: Multi-turn.
LLM with RAG and a vector database. Technology used: Large Language Models, RAG architecture, Vector database, Prompt engineering, Document processing, Context window optimisation.
Technology used
Talk to the people who built this
theBlue.ai comes back within one business day.
