Company Cortex is an AI knowledge base that turns your documents, policies, and procedures into quick, source-backed answers for staff and customers.
Six practical ways to make approved knowledge easier to find and use.
Ask a question in plain English and get a sourced answer quickly. No more digging through folders or waiting for colleagues.
Approved documents can be re-indexed after updates, with freshness depending on the configured source and sync schedule.
We agree hosting, model access, roles, logging and retention around the sensitivity of your source material.
Supported PDFs, documents, spreadsheets, web pages and CRM notes can be indexed within agreed access and format boundaries.
New starters query the knowledge base like a patient, expert colleague. Guided answers help them find policies, processes, and product information sooner.
Deploy a customer-facing version for supported common questions, with clear escalation when a human should step in.
At its core, Company Cortex uses a technique called GraphRAG (Graph-based Retrieval Augmented Generation). In plain English, the system searches approved indexed documents for relevant passages before constructing an answer. Responses can cite their sources so the user can check where the answer came from.
Sources
Cited in answers
Roles
Access controlled
Logs
Auditable queries
An AI answer can sound confident even when it is wrong. This is often called a hallucination. Source-backed retrieval can reduce that risk by using approved indexed material. We also define when the system should say it cannot find enough evidence or hand the question to a person.
“The difference is simple: the answer can point back to the source.”
Before relying on a company knowledge system, pick a small set of real questions. Include one with an approved answer, one where two documents disagree, and one that the source material cannot answer. For each response, check the cited passage and document version, whether the person asking has permission to see it, and whether the system stops or hands over when the evidence is missing.
Record the expected answer beside the observed one. A citation alone is not proof of accuracy: the passage might be outdated, incomplete or about a different case. That check gives your team a concrete basis for approving sources, access and answer boundaries.
Suppose an old visitor guide says to use the east entrance, while the current approved version says the west entrance. An answer that cites the old guide accurately is still wrong for today. Check both the answer and the version it used.
Our synthetic test pack supplies fictional documents and six checks: a current answer, conflicting sources, missing information, restricted access, permitted access and a source update. It includes expected behaviour and a blank worksheet for your observations. These are teaching examples, not completed Company Cortex tests or client results.
Run the checks in an isolated test environment. A handful of correct answers cannot establish overall accuracy or security. If a restricted answer leaks, fix it and retest before releasing affected access; other correct answers do not cancel that failure.
“A plausible answer is difficult to check when it does not identify the source or explain what is uncertain.”
→ More work for the person checking the answer
“Retrieves passages from approved indexed documents, shows the supporting source and can be configured to stop when evidence is missing.”
→ Source-backed answers with a reviewable trail
Your business data needs explicit boundaries. We agree the deployment, encryption, model access, user roles, retention and query logging before launch. The final controls depend on the chosen infrastructure and source systems, then we test those controls against realistic user roles for compliance and audit purposes.
Internal staff tool: Your team asks questions like "What's our returns policy for damaged goods?" or "How do I process a BACS refund?" and receives an answer from the approved knowledge set, with a link to the source document and the configured access rules applied.
Customer-facing FAQ bot: Embed Cortex on your website and let customers find answers to common questions without waiting for a support agent. Product specifications, delivery timescales, warranty terms — available on demand from the approved knowledge set.
Onboarding accelerator: New starters query Cortex as if it were a patient, knowledgeable colleague. Company processes, software guides, HR policies — everything a new hire needs to become productive, available on demand from day one.
Company Cortex can connect with Amplio, our unified communications platform. This means your AI assistants can pull verified answers from Cortex during live customer conversations across email, SMS, WhatsApp, and voice. The result is a more consistent, reviewable first response, with human escalation where needed.
A structured four-phase approach from audit to ongoing refinement.
We map your existing knowledge assets — documents, policies, procedures, FAQs, CRM notes — and identify gaps and priorities.
We prepare the agreed documents for search, checking formats, access and extraction quality before indexing. Supported formats can include PDF, Word, Excel and HTML.
We configure retrieval parameters, test accuracy against real-world questions, and refine until answers are consistently reliable.
Deploy to your team (and optionally customers), monitor usage patterns, and continuously improve accuracy based on real queries.
We review the source material, access rules, answer boundaries and deployment needs before quoting the build and any ongoing support.
Hosting, model, storage and connector costs are itemised before work starts.
Security depends on the deployment and integrations we agree. Before we build, we document the data flow, hosting, encryption, model access, user roles, retention and logging, then test the agreed controls. Without that scope, we won't make an absolute security or compliance promise.
The scope can include PDF, Microsoft Word, Excel, PowerPoint, plain text, HTML and CSV, plus supported web pages, CRM records or email archives. We check sample files, access permissions and extraction quality before confirming which sources can be indexed.
Company Cortex uses retrieval-augmented generation to answer from approved indexed documents rather than model memory alone. Responses can include source citations. We test realistic questions, tune retrieval and define when the system should stop or hand over; no AI system is perfectly accurate.
Updated documents can be re-indexed, and supported sources such as SharePoint, Google Drive or a CRM can use an agreed sync schedule. Freshness still depends on source access, sync health and the documents your team approves.
Yes. A customer-facing version can use its own access controls, curated knowledge set and human-handoff rules. It can answer supported product, delivery and policy questions and show which enquiries still need a person.
Before you commission the work
Choose a small set of questions staff repeatedly ask, then locate the approved answers. Record who owns each source, when it was checked and who may access it. Conflicting or outdated documents need attention before retrieval can be trusted. Test whether the system can identify an unsupported question and direct it to a person, as well as answer familiar ones.
Tell us what's slowing you down. We'll help you find the first targeted improvement.