Controlled Interactive AI Manual
Give machine users contextual answers from approved sources, with a citation for every answer and visual procedures for the work itself
Our role
We designed the machine-context model, controlled retrieval and citation architecture, role and safety rules, visual-procedure integration, and expert evaluation framework.
Competencies involved
Case images
See the concrete visuals, test environments, and outputs that make the story easier to assess.
The Problem
Many companies face similar challenges when trying to create value from their technical data and processes.
Operators and technicians need answers in the context of the exact machine they are using, but documentation is difficult to search and generic AI can blend versions, roles, or unsafe advice. A useful assistant must know when to answer, what evidence applies, and when to stop.
These challenges often result in concrete problems:
- Search results do not reliably account for machine, serial, option, software, and document-version context
- Uncited or cross-version answers are unsuitable for operation, maintenance, and safety-sensitive decisions
- Text answers alone cannot replace the visual sequence and confirmation points required for physical procedures
Why It Matters
Why does this matter? Look at the broader impact:
An interactive manual only earns trust when its answers remain inside approved evidence and role boundaries. Safety restrictions, escalation, and expert-tested behavior are product requirements, not additions after the AI works.
Key metrics we focus on:
Relevance of retrieved evidence to machine, serial, configuration, and version context
Citation coverage, grounded-answer quality, and correct refusal or escalation behavior
Usability of the transition from a question to the applicable visual procedure
Our Solution
Here's how we approach the solution:
Controlled retrieval over approved manuals, service bulletins, and procedures filtered by machine, serial, configuration, version, and user role
Answer contract requiring citations to supporting passages, explicit limits, and refusal when approved evidence is absent or conflicting
Safety and escalation layer that blocks restricted guidance, separates operator and technician actions, and routes unresolved cases to experts
Integration with interactive visual procedures plus an expert evaluation set covering expected answers, evidence, unsafe prompts, and edge cases
Results
The comparison between the starting point and the result shows the concrete value the solution can create.
Users search documents manually or ask general AI without reliable machine context, citations, or safety boundaries
Context-aware manual experience that answers from approved evidence and opens the visual procedure for guided execution
The concrete results include:
- Users reach applicable information through one interface while retaining visible evidence for every answer
- Role, safety, and escalation rules keep the assistant inside the responsibilities defined by the manufacturer
- Expert evaluation makes retrieval gaps and unsafe behavior visible before controlled release
Project context
An equipment manufacturer wanted to make complex manuals easier to use without introducing an unrestricted chatbot. neexo scoped approved sources, defined machine and user context, connected answer retrieval to visual procedures, and built an evaluation set with domain experts to govern release decisions.
Deliverables
- Machine-context retrieval pipeline with approved-source, version, serial, configuration, and role filters
- Citation-first interactive manual with refusal, safety restriction, escalation, and visual-procedure handoff
- Expert evaluation set, scoring rubric, reviewed failures, and controlled release criteria
Want an AI manual that knows the machine, cites its sources, and respects safety boundaries?
We can help analyze your situation and identify opportunities for similar solutions.