Skip to content
Abdullah Ghonaiem
All projects

AI & Automation

Internal Knowledge Assistant

A RAG chatbot that answers employees’ questions from company records and scanned documents.

Internship
Diyar United Company
My role
AI Engineer Intern
Period
2023
Key result
RAG + OCR answers grounded in internal documents
  • Python
  • RAG
  • OCR
  • LLM Integration
  • Document Processing

Overview

During an internship at Diyar United Company, I developed an internal chatbot in Python. Employees ask a question in plain language, about employee information or internal documentation, and the assistant answers from the company’s own material.

It was built with retrieval-augmented generation: instead of relying on what a language model already knows, the assistant first finds the relevant passages in a knowledge base and then answers from them.

Problem and context

Useful internal information was spread across employee records and organizational documents. Finding an answer meant knowing which document held it and reading until you found the line.

Many of those documents were not searchable text at all, so a search box would not have helped. And a general-purpose language model knows nothing about one company’s people or policies, so its answers had to be tied to the source material.

My contributions

Designed and developed

My own work

  • The Python chatbot and its question-answering flow.
  • OCR processing to extract text from organizational documents.
  • A searchable knowledge base combining employee records and company documents.
  • Retrieval that supplies the relevant passages to a large language model for each question.

Done with the team

Shared work

  • Developed as an intern, within the company’s team and on its internal data.

Technical architecture

From internal documents to an answer

Conceptual diagram

  1. 01

    Sources

    Employee records · documents

  2. 02

    OCR

    Text extracted from documents

  3. 03

    Knowledge base

    Searchable content

  4. 04

    Retrieval

    Passages relevant to the question

  5. 05

    Answer

    Written by an LLM from that context

Conceptual retrieval-augmented generation flow. It does not name the specific models or storage used.
Ingestion
Documents pass through OCR where needed, and the extracted text joins the employee records in one knowledge base.
Retrieval
For each question, the system searches the knowledge base for the passages most relevant to it.
Generation
The question and the retrieved passages go to a large language model, which writes an answer from that context.

Engineering challenges

  1. Documents that were not text

    Challenge
    Much of the organizational material could not be searched or read by software, so it was invisible to any retrieval system.
    Approach
    I added an OCR step that extracts the text from documents before they enter the knowledge base.
    Technology
    • Python
    • OCR
    • Document processing
    Outcome
    Document contents became available for the assistant to retrieve.

  2. Answers about one specific company

    Challenge
    A language model on its own cannot know an organization’s employees or internal documentation, and will produce a plausible answer anyway.
    Approach
    Using retrieval-augmented generation, each question is answered from passages retrieved out of the company’s own records and documents, supplied to the model as context.
    Technology
    • RAG
    • Information retrieval
    • Large language models
    Outcome
    Employees could ask natural-language questions and get answers based on internal sources.

  3. Two kinds of source material

    Challenge
    Structured employee records and free-form documents are stored and read differently, but a person asking a question does not care which one holds the answer.
    Approach
    Both were integrated into a single searchable knowledge base, so one question can draw on either.
    Technology
    • Python
    • Knowledge base
    • Data integration
    Outcome
    One interface answers questions about both people and documentation.

Results and impact

  • Working internal chatbot for natural-language questions
  • Document contents extracted and made retrievable
  • Employee records and company documents in one knowledge base

Visuals

No product screenshots published

Internal company system. Screenshots are intentionally withheld; the diagram above is a conceptual view. The cover and the architecture diagram on this page are conceptual illustrations.