Skip to content
All work
Working prototypeAI & Internal Tools

Internal knowledge assistant

An assistant that answers staff questions from a small library of approved policies, citing the source for every answer and admitting when it does not know.

Business
Fictional distribution company
Status
Working prototype
Example project. The business, names, data and figures shown here are fictional.
How many vacation days in year one?
Full-time staff receive 10 days in their first year…Source · Leave Policy §2.1

01 · Business problem

The problem

Staff ask the same questions about expenses, leave, returns and safety procedures. The answers exist in documents nobody can find quickly, so experienced staff field the questions instead.

Proposed approach

The assistant searches only the approved documents, answers with the relevant passage and cites it. If nothing relevant is found, it says so and logs the question so the documentation can be improved.

02 · System architecture

How it fits together

  1. 01

    Document library

    Approved policies and SOPs, each with an owner.

  2. 02

    Retrieval

    Finds the most relevant passages for each question.

  3. 03

    Answer

    Responds from those passages with citations.

  4. 04

    Feedback log

    Tracks unanswered questions and helpfulness ratings.

03 · Proposed workflow

Step by step

  1. 1

    Ask

    Employee asks in plain language.

  2. 2

    Retrieve

    Relevant passages found in approved documents.

  3. 3

    Answer

    Response with source citations.

  4. 4

    Learn

    Gaps logged for document owners.

04 · Interactive demo

Ask the policy assistant

Ask about a fictional company’s policies. Every answer cites its source, and it says so when it doesn’t know.

Policy assistant · Maple & Pine Supply Co. (fictional)Demo · fictional data

Ask a question about company policy.

Try one of these, or type your own:

Simulated: answers come from keyword search over six fictional documents in your browser. No AI model is used and it has no access to any real company’s systems.

05 · Scope

Deliverables

  • Document ingestion
  • Grounded answers with citations
  • Unanswered-question log
  • Feedback capture

This prototype is built with

  • Next.js
  • TypeScript
  • Keyword retrieval running in the browser
  • Fictional policy documents

Likely production stack (proposed)

  • Embedding-based search
  • A large language model on a secure server
  • Slack or Teams app
  • SharePoint or Drive connector

Results

No results reported. This is a working prototype, not a live deployment, so there are no real measurements to share.

Related service

Internal Knowledge-Base Assistant

Next project

Sales follow-up drafting

Want something like this for your business?

We’ll talk through how it would work with your tools, your data and your team, and be honest about whether it’s worth building.

Book a discovery call