Connect.Understand.Apply.Operate.
SALTIR brings documents, master, machine and process data together and builds AI applications on top for production, administration and knowledge management. Accepted against a specification and operated by us.
Not a workshop. An accepted solution.
Many AI projects end with slides. SALTIR ends with a system that was accepted against a specification and runs in operation.
What is often sold:
AI workshop with a list of ideasWhat you get with SALTIR: SpecificationTask, data sources and measurable acceptance criteria, agreed in writing.
What is often sold:
Data strategy on slidesWhat you get with SALTIR: Delivery through acceptanceWe build, integrate and test against the agreed criteria.
What is often sold:
Proof of concept without operationWhat you get with SALTIR: Operation and maintenanceMonitoring, updates and model care as a service, with one fixed contact.
What is often sold:
Lock-in to one platformWhat you get with SALTIR: Technology-openYou choose the model, operating site and data sources. Every module stays replaceable.
The first visible benefit: an orderly file system.
SINTER reads every file and understands its content. The result is a proposal for consistent folders, file names and descriptions.
- Propose before changing
- SINTER creates a plan. Changes are applied after approval.
- Nothing gets lost
- Every change is logged and reversible. Old paths remain traceable.
- Rules, not one-offs
- The scheme is defined once and applies to every new file as well.
- Permissions stay intact
- Access rights are inherited from the source, not reassigned.
- Customers_old/Quote final.pdfRev. A · superseded
- CUSTOMERS 2021/Quote_final_v2_NEW.pdfRev. B · valid
- Cust. docs/QT-2023-114 (copy).pdfDuplicate
- Projects/SB2041 test plan new.docxRev. C · valid
Five building blocks. One per data type.
SINTER
Turns scattered document collections into an organized, versioned and citable knowledge base. The LECTIO stage ingests documents in structured form.
- Data type
- Documents
- Same thing
- Copy, version, translation, redundant passage
- Origin
- File, location, page, position
- Human review
- Review list for versions and contradictions
- Capabilities
- Read
- Link
- Enrich
- Reconcile
- Index
- Query
- Organize files
CURSUS · Example with synthetic sample data
Where does the time go?
Overhauling a large plant is planned for 14 months and overruns almost every time. From 18 years of ERP exports, CURSUS reconstructs the actual process and breaks the delay down into its causes, from the data alone.
- 29,033
- Events from 3,026 orders
- 44
- Variants instead of one process
- +2.6
- Months of average delay per period
- 65%Waiting for special procurement
- 27%Approval loops for additional scope
- 8%Rework
Procurement bottleneck
Specially procured parts wait a median of 34 days. For two material groups, 9 days recently passed before an order was even placed.
Approval loops
Additional scope takes an average of 2.4 attempts to get approved. Each rejection round costs about a week.
Predictable finding
The same finding on the same assembly occurs in 11 of 14 periods. With it, a period overruns by 3.1 months on average; without it, by 0.6. It belongs in the plan, not in the delay statistics.
Plan vs. actual
Planned scope follows the target process in 87% of cases, additional scope in none. That is where the delay comes from.
One order. All data in one place.
A fault on the packaging line. SALTIR brings together what the service team needs, from every system, right at the machine.
- SINTERValid manual and wiring diagram
- SPINESpare part and stock level
- PLEXUSHistory of the machine data
- CURSUSEarlier faults and how long they took
- LOQUIThe answer, read out at the machine
How the building blocks work together.
- SINTERSPINE
Facts from documents are attached to SPINE objects. A contradiction between datasheet and master data is flagged automatically.
- PLEXUSSPINE
Machines, sensors and signals are objects in SPINE. Signal register and ontology describe the same plant.
- PLEXUSSINTER
A fault message points to the valid manual and the valid schematic of the affected machine.
- SINTERCURSUS
Dated documents such as repair orders and test reports become events. This creates a process even where the ERP system has gaps.
- PLEXUSCURSUS
Downtime, setup and faults become events in the same case. Office process and machine appear in one flow, with the case objects coming from SPINE.
- LOQUIAll
Speech as input and output: the technician asks at the machine and hears the answer read aloud.
Built in layers.
- Connectivity
- One connector concept for all building blocks: source, credentials, mapping and permissions per location.
- Review
- One shared review list. Uncertain results are flagged, not waved through.
- Measurement
- One test bench per building block with test corpus and error rate. A prerequisite for every commitment to customers.
- Model gateway
- One shared access point to language models, with a model per task, local or in the cloud.
- Sovereignty & Operations
- Data stays in-house: operated at SB or on the customer's premises, verified data paths, European providers.
- Document co-pilot
- Invoice checking
- Meeting notes
- Data products
- SINTER
- SPINE
- PLEXUS
- CURSUS
- LOQUI
- Connectivity
- Review
- Measurement
- Platform
- Model gateway
- Sovereignty & Operations
Technology-open. Every module replaceable.
Specification, data and acceptance stay fixed. You choose the model, operating site and data sources to match your requirements, and you can swap them later.
- Stays fixedSALTIR modulesSINTER · SPINE · PLEXUS · CURSUS · LOQUI
- Always stays in GermanyDevelopment, integration, operation, support
- Language modelOpen source, local
- Operating siteEdge
- Data sourcesERP & CRM
Open source, local: Runs entirely in your infrastructure, no data leaves the building.
A starting point with a fixed result.
- Fixed price · Weeks 1–4
AI Decision and Risk Check
Assess context, impact and risk. Result: a rated portfolio of use cases.
- Weeks 5–17
90-Day Pilot
One use case, real data and a clear go or no-go at the end.
- Next
MVP & Rollout
Industrialization and integration into ongoing operations.
- Ongoing
Operations & Service
Maintenance and lifecycle of model and hardware.
From the pilot onward, our engineers work directly on site, from sensors to model.
What is in your data? Let us find out.
We look at a sample of your documents, master, machine or process data and assess what can be done with it.
Where does your operation lose time? We find it in your data.
Orders, records and machine data show where processes stall and what that costs.
What should your AI solve? We test it on your data.
We assess your use case on real data and tell you plainly whether it is worth it.