AI technical supportfor medical devices.
Manuals, spare parts and device status in one conversation. Two MCP servers connect technical knowledge and operational data to support technicians, workshops and warehouse teams.
Which procedure and parts support this service task?
Technical knowledge
Search · Fetch
Operational data
Inventory · Work records
One connected view
Technical answers meet operational data.
Managing ventilators and home-therapy equipment requires coordination between field technicians, internal support, the workshop and the warehouse. PDF manuals, Excel files, local systems and people’s experience each hold part of the answer. Handling a reported issue requires both the technical procedure and current information about the serial number, spare parts and work in progress. The project connected these dimensions in a shared support experience.
- Agentic AI
- Vector store
- MCP and operational data
What we designed.
The project calls for more than searching manuals. Documented knowledge changes as procedures evolve, while inventory and work progress evolve throughout the day. Two MCP servers separate these domains and make them usable in the same conversation, connecting the answer to technical sources and operational context.
Structure technical knowledge
Service manuals, protocols, fault tables and procedures were analysed and segmented into searchable content. Procedures, components, error codes and tests were organised in a semantically indexed technical vector store linked to the original documentation.
Develop two MCP servers
The first server exposes search and fetch tools to query technical knowledge and retrieve original content. The second connects master records, stock, serial numbers, movements and work status. AI models can access these complementary sources through dedicated tools.
Unify support and operational context
The conversational layer brings together the information relevant to the question: the procedure, component availability and the device’s status. The system was tested on technical support, spare-parts searches, serial-number checks, inventory and work-planning scenarios.
Manuals, serial numbers, parts and work records
Vector store + two MCP servers + AI models
Contextual technical support
The right procedure,in the real context.
Knowing the required procedure is only part of the answer. Available components, serial-number status and work already performed on the device matter too.
Three information domains to connect.
Technical knowledge
The document collection provides material for understanding reported issues and consulting procedures.
- Service manuals and protocols
- Error codes and components
- Procedures and inspection tests
Items and spare parts
Inventory data connects technical knowledge with the resources available.
- Master records and product families
- Stock levels and coverage
- Movement history
Serial numbers and work records
Each device is considered in the context of its history and the workshop workflow.
- Serial-number status
- Workshop arrivals and departures
- Work progress
Two knowledge sources.One operational assistant.
Explore how documents and process data become accessible in the same conversation.
Make technical knowledge searchable.
PDF content is segmented and indexed. Semantic search identifies relevant information; source access allows further investigation.
- Manuals and procedures
- Components and error codes
- Original pages and sections
A technical vector store linked to documentation.
Connect the answer to what is happening.
Stock, serial numbers and work progress bring operational context into the conversation. Technical knowledge can be read alongside device status.
- Items, spare parts and stock
- Serial numbers and work records
- Movements and product families
Information updated through connected operational systems.
Give models distinct, complementary access.
One MCP server exposes technical knowledge; the other exposes operational data. The assistant uses the tools relevant to the question and connects retrieved information.
- search · Semantic retrieval
- fetch · Source access
- Inventory and work data
An integration layer between models, documents and data.
Bring the procedure and the real situation together.
A question may call for several sources: relevant documentation, an available part and serial-number status. The experience combines them to support people’s work.
- Technicians and internal support
- Workshop and warehouse
- Process managers
Contextual support tested on work scenarios.
The structure behind an operational assistant.
Document preparation, data access and testing are distinct parts of the same development effort.
From PDFs to a technical vector store
Work starts by turning documentation into a searchable knowledge base.
- Analysis and segmentation
- Manuals, tables and procedures are split into content relating to components, error codes, spare parts and checks.
- Semantic indexing
- The vector store supports retrieval based on the meaning of a question, beyond matching individual words.
- Original documents
- Source retrieval provides access to pages, sections and instructions in the technical documentation.
MCP for documentation and operations
Two servers provide models with tools that have different responsibilities.
- Documentation server · search
- Semantic search across procedures, components and error codes.
- Documentation server · fetch
- Retrieval of original content to examine and verify the information.
- Operations server
- Access to item, stock, serial-number, movement and work data, updated through the connected systems.
- AI models
- The MCP layer connects documented knowledge and operational information to interfaces and models such as ChatGPT and Claude through the available tools.
One experience for different roles
Questions change with each user’s work; the sources remain connected.
- Field technicians
- Retrieval of relevant documentation and the device’s technical and operational context.
- Support and sales teams
- Access to information that helps them handle a reported issue without reconstructing it through informal exchanges.
- Workshop
- Access to serial-number status, work progress and the information needed to manage the workflow.
- Managers and warehouse teams
- Visibility into availability, coverage, movements and requirements to support operational decisions.
Testing on work scenarios
The assistant was tested on technical and operational questions, including requests spanning multiple sources.
- Technical scenarios
- Procedure searches, error-code interpretation and identification of components and spare parts in the documentation.
- Operational scenarios
- Serial-number checks, stock and coverage analysis, workshop workload and historical spare-parts usage trends.
- Evaluation criterion
- Checking response consistency and the ability to connect technical knowledge with operational data in the context of the question.
An infrastructure for the next generation of services.
The infrastructure can be extended with an app for field technicians, a workshop dashboard, spare-parts alerts and planning agents. These extensions require dedicated development and validation, building on the established connection between technical knowledge and operational systems.
The assistant supports information retrieval and technical decisions. Checks, interventions and testing remain the responsibility of qualified personnel following the applicable procedures.
Questions that connect different sources
- Which documentation is relevant to this error code and model?
- Which spare parts are available for the planned activity?
- What is the work status of this serial number?
- Which stock movements and coverage levels apply to this product family?
What it makespossible.
- Structured, searchable technical knowledge with access to original documents.
- Combined access to procedures, spare-parts availability and serial-number status.
- A shared entry point for technicians, workshops, support teams and operations managers.
- Assistant evaluation across technical, logistics and work-management scenarios.
Does your support team still search for answers across manuals, business systems and colleagues?
Does your support team still search for answers across manuals, business systems and colleagues?
Tell us about your project
AI to accelerate transcriptomic research.
Geneversity: millions of RNA profiles to identify cells, estimate their proportions and explore new biomarkers.