A maintained knowledge collection
Ingestion and indexing for agreed documents or records, with source identity, update handling and ownership. We consider how corrections, deleted content and superseded versions should affect what can be retrieved.
Knowledge systems / Answers with context
Kerning / IntelligenceAI knowledge assistant development
Help people find and work with information from the sources they are permitted to use. We build knowledge assistants around a defined collection, useful questions and a clear relationship between an answer and its supporting material.
Finding an answer is only part of the job.
Information can be spread across documents, records and conversations, with different owners and access rules. A convincing answer is not enough if the reader cannot identify its source, recognise an outdated policy or understand when the available material does not resolve the question.
We define the assistant’s remit before choosing the interface. That includes which sources it may use, how those sources are maintained and which questions should result in a reference, a clarification or a clear admission that the answer is unavailable.
01Scope / Shaped around your process
What we can build.
Permission-aware retrieval
Retrieval designed around the user’s permitted information, including boundaries between teams or organisations where needed. Access rules are part of the data workflow, not only the visible chat interface.
Answers with source context
An interface that can connect an answer to supporting material, surface relevant excerpts and ask for clarification. The response behaviour includes questions the available knowledge cannot answer reliably.
Evaluation and feedback tools
A representative question set and review process covering factual support, source selection and unanswered questions. Feedback provides a route to improve the collection or assistant behaviour without treating every response as authoritative.
02Delivery / From context to working software
How the work takes shape.
Start with questions and sources
Identify the users, recurring questions and authoritative materials. Review document quality, ownership and access before deciding which collection is suitable for a first release.
Build retrieval before expanding scope
Check whether the system can locate the right material for the agreed questions. Then evaluate the answer format, references and uncertainty behaviour using examples the source owners can assess.
Keep the knowledge accountable
Define source updates, review responsibilities and how users report an issue. If actions are later added, treat their permissions and approvals as a separate workflow decision rather than an automatic extension of search.
Context / Relevant work
Conversational assistance in defined contexts.
The BAT planning prototype includes a planner copilot connected to structured prototype data and analysis. Cronuss AI is a separate application in development covering legal research, document review, drafting and collaboration. These projects show distinct contexts for working with information.
BAT uses mock data and has no demonstrated live SAP, POS or Pack Smart integration. Cronuss AI is described as development work, not a deployed knowledge-assistant result.
03Before you begin
Useful questions.
How is this different from a general chatbot?
The assistant has a defined collection, user permissions and a specific purpose. Its answers are evaluated against those sources and questions. A general conversational interface alone does not establish that connection or maintain the underlying knowledge.
Can it answer only from approved sources?
Source selection and response rules can be designed around an approved collection, with references and explicit handling of missing answers. Those behaviours still need evaluation; limiting the source set is not a guarantee that every generated answer will be correct.
Can different users access different information?
Yes, where the source systems and identity model support the required rules. We define and test permission boundaries through ingestion, retrieval and response generation, including what happens when a user’s access changes.
What happens when our documents change?
The scope should define update frequency, ownership, deletion and version handling. An assistant is only as useful as the information maintained for it, so source operations and ongoing review belong in the delivery plan.
Next / Your first conversation
Start with the questions your team keeps asking.
Bring a small set of representative questions and the documents that should answer them. That is a practical basis for defining the assistant’s scope and evaluating its usefulness.
Discuss a knowledge assistant