Skip to content

Knowledge DB and Chatbot and Helpdesk self-service

Intro

If you wanna manage knowledge in our company, so that

  • the Role Role: Customer
    in a App: Helpdesk
    can
    • use Capability:search and find content at anytime
    • use a Capability:chatbot and ask questions
  • the Role Role: Support employee
    in a App: knowledgeDB
    can
    • add content as Template: runbook and Template: FAQ to Capability: KnowledgeDB, as part of

      "ITILs Closure: Once the incident is resolved and the service is restored, the incident is closed. "
  • the Role Role: Developer
    in an App: knowledgeDB
    can
    • add documentation in Template:service_markdown

Option1: Handcrafted solution. use Confluence as knowledgeDB

One can manage the knowledge data in Confluence.

But the extraction of the unnecessarily complex.

  • Thats the path only for the Capability:chatbot
  • The problem is - with the unstructured content on confluence.

Network

https://www.youtube.com/watch?v=zjdFOwBxBRo

Option2: Handcrafted solution. Use Git as Knoledge DB

One can manage the knowledge data in Git, in Markdown.

https://medium.com/@dinber19/creating-your-own-q-a-chatbot-over-your-markdown-documents-using-langchain-11645bc346bc

This is how to works

Configure

- choose text-embedding-ada-002 is a text embedding model developed by OpenAI. It's designed to convert text into numerical representations (vectors) that capture the semantic meaning of the text. These vectors can then be used for various natural language processing tasks, such as search and chat - choose a vector DB "Weaviate" to store the data, ready for queries

Load and Split documents and generate embeddings and store

- use Unstructured library implemented in LangChain.
LangChain facilitating interaction between LLMs and the environment - the result is stored in vector DB Weaviate

#### Query the LLM usingthe vector DB - use AzureOpenAI to query

GUI

- use "Azure AI Bot Service" https://learn.microsoft.com/en-us/azure/bot-service/bot-service-design-pattern-embed-web-site?view=azure-bot-service-4.0 - Embed bot into the documentation page https://github.com/squidfunk/mkdocs-material/issues/1804 - Connect a bot to Web Chat https://learn.microsoft.com/de-de/azure/bot-service/bot-service-channel-connect-webchat?view=azure-bot-service-4.0#get-your-bot-secret-key

Network

Advantage: the linting of the markdown gives one the quality assurance

  # Runbook Title

  ## Overview
  Provide a brief description of the runbook's purpose and scope.

  ## Prerequisites
  List any prerequisites or requirements needed before executing the runbook.

  ## Steps
  1. **Step 1: Description**
    - Detailed instructions for Step 1.
    - Include any necessary commands or code snippets.

  2. **Step 2: Description**
    - Detailed instructions for Step 2.
    - Include any necessary commands or code snippets.

  3. **Step 3: Description**
    - Detailed instructions for Step 3.
    - Include any necessary commands or code snippets.

  ## Troubleshooting
  Provide common issues and their solutions.

  ## Contact Information
  List the contact information for support or escalation.

  ## Version Control
  - **Version:** 1.0
  - **Last Updated:** YYYY-MM-DD
  - **Updated By:** Name

  ## References
  Include any references or links to additional resources.

Cost: Free to low, for AI queries. Like 15$ for 1 Million tokens https://azure.microsoft.com/en-us/pricing/details/cognitive-services/openai-service/

Option3: Some Buy solution like zendesk.de

Like https://www.zendesk.de/

  • Messaging and live chat
  • AI and automation
  • Data privacy and protection
  • Help center
  • Agent workspace
  • VoiceNew
  • Workforce engagement

Disadvantage: data lives outside the company

Cost: 55EUR per 2nd level per Month. like 3000 EUR yearly https://www.zendesk.de/pricing/

Option4: Use the "Gemini kill RAG" alternative

Google Just Killed RAG 🤯
I just parsed 6,000 pages with >99% accuracy—for just $1.
https://www.linkedin.com/feed/update/urn:li:activity:7295488197940166656/

But teh DB is still needed

``` “what’s the lifespan of the context cache”? As far as I can tell, it’s only a few hours at best. ````

But there is a solution, which actually replaces the load, split, embed part

  • Use Gemini 2.0 Flash to convert PDF pages directly into chunked text.
  • Store chunks in KDB.AI for vector search.
  • Tie it all together in a RAG workflow.

See https://ai.gopubby.com/10x-cheaper-pdf-processing-ingesting-and-rag-on-millions-of-documents-with-gemini-2-0-flash-8a93dbbb3b54