How a Professional Services Company Built a Searchable Internal Knowledge Assistant

A Beirut professional services company kept policies, manuals and templates across many folders. We built an internal knowledge assistant with indexing, citations, permission-aware search, gap tracking and an update workflow.

Project overview

A professional services company in Beirut kept procedures, manuals and templates in many folders, so employees asked colleagues where things were. We built an internal assistant that answers from approved documents. Administrators stay in control of the content and who can see it.

The solution

We implemented five connected automations:

Indexing and categorization of approved documents
Cited answers to questions asked in plain language
Permission-aware retrieval that follows existing access rights
Gap tracking of questions the documents do not answer
Re-indexing and escalation to a named person when the documents are not enough

Business impact

Employees now find internal information through one assistant.

Answers come from approved documents, with a link to the source
Restricted documents stay restricted
Questions the documentation does not answer are logged for administrators

The challenge

Employees needed to know which folder held a procedure and which version was current. When they did not, they interrupted a colleague who did, and the answer depended on who happened to be asked.

Outdated documents sat next to current ones. Nobody could see which questions kept coming up, which documents were never used, or where the documentation had gaps. Some material was restricted to certain teams, so a single shared search could not simply be opened to everyone.

Individual automations

The parts fall into two groups: those that prepare the knowledge, and those that serve employees and keep it healthy.

Ingestion and indexing of approved documents

The assistant only knows what the company has approved for it.

Administrators select the document collections that may be used: policies, procedures, manuals, templates and internal documents.
The workflow reads each document and splits it into passages.
The passages are indexed so that they can be found by meaning, not only by keyword.
Each passage keeps a link to its source document and location.

Only collections that administrators have connected and approved are indexed. Nothing is added automatically.

Document categorization and metadata

Documents carry the information that makes answers trustworthy.

Each document is assigned a category, an owner, a version and a review date.
The assistant can prefer current versions and avoid superseded ones.
Categories also let employees narrow a search, for example to one department.

Owners and administrators maintain the metadata and can correct it.

Natural-language search with cited answers

Employees ask questions the way they would ask a colleague.

The employee types a question in the conversational interface.
The assistant retrieves the passages that are most relevant.
It writes an answer based on those passages only.
The answer shows citations or links to the source documents, so the employee can open the original.

The assistant answers only from retrieved documents. It does not use information that has not been indexed.

Permission-aware retrieval

People only receive answers from documents they are allowed to see.

The assistant respects the access controls that already apply to the documents, where those controls have been configured.
A restricted document is never retrieved for someone without access to it.
Administrators can review which collections each group of employees can search.

Access is decided by the company's existing permissions, and administrators remain responsible for them.

Content-gap and unanswered-question tracking

Questions the assistant cannot answer are valuable information.

Questions with no suitable source are recorded, without the employee's private details being exposed to other users.
Repeated questions are grouped, so administrators can see which topics are missing from the documentation.
A short report lists the gaps, ordered by how often they occur.

Administrators decide which gaps deserve a new or updated document.

Document update and re-indexing workflow

The index stays in step with the documents.

When a document is added, changed or removed in a connected collection, the workflow detects it.
The changed document is re-read and re-indexed, and removed documents are taken out of the index.
Documents that pass their review date are flagged to the owner.
The date of the last update is shown for each source.

Administrators can remove a document from the assistant at any time.

Escalation when information is missing, outdated or inconclusive

The assistant is built to stop rather than guess.

If no suitable source exists, the assistant says so.
If sources disagree, or the only source is out of date, it says that and shows the sources.
In each case it names the appropriate person or team to ask, using the contact list the company provides.

A person always remains the final source when the documents do not give a reliable answer.

How the workflow operates

This is what happens when an employee asks a question.

An employee asks a question The question is typed in plain language in the conversational interface.
Access is checked The assistant works out which approved collections the employee is allowed to search.
Relevant passages are retrieved The index returns the passages that best match the question, preferring current versions.
An answer is written from the sources The answer is based only on the retrieved passages and shows the citations.
The employee checks the source The employee opens the cited document to confirm the answer or to read more.
Gaps are recorded If no reliable source exists, the question is logged as a gap and the employee is told whom to ask.
Administrators keep it current Gap reports and review dates tell administrators what to add, update or remove, and changed documents are re-indexed.

Systems and integrations

The assistant is connected only to what has been configured for the company.

Document collections: the approved folders or libraries that administrators connect for indexing.
Access controls: the company's existing permissions, applied to retrieval where they have been configured.
Contact directory: the people and teams named when the assistant escalates.
Admin portal: manages collections, categories, owners, review dates and the gap report.

Related services

AI Agents

All insights · Get in touch