05 Web Portal in General

The web portal can be reached at the following web address: https://[YOUR PORTAL URL].appspot.com/. The web portal is a platform on which various customer-specific processes and assets are integrated. The process and measurement data are displayed and processed in different ways, e.g., as tiles for an overview, cockpit views of live data and individual analyses so that the organization can successfully draw conclusions about the current status and added value.

05.1 Platform Structure

The figure below shows the structure of the portal pages and their link points. The individual pages can be viewed and edited depending on the authorization. In the following, all the different pages and their functioning are explained. The illustration serves for orientation on the platform.

05.2 Login

After the administrator has added a new user, the user receives an e-mail with his personal password. With this password and the e-mail address, it is possible to log in to the portal. The user can also login with Google credentials and click the 'Forgot password' button to receive an email to reset the password. The login window of the scitis portal can be reached via the following link. https://[YOUR PORTAL ADDRESS].appspot.com/

The login page is not available when using Google Single sign-on!

05.3 Header and logout

The logged in user will see the account name in the upper right corner. After clicking on this name a dropdown menu opens. Here there are the areas Settings and Support, which, depending on the user authorization, forward to the corresponding portal settings. The pen button next to the user's email address takes you to the user's profile settings. The Logout button logs out the current user.

The functions in the Settings paragraph are used to manage all functions, users and assets of the portal and is divided into the following pages:

• Users - add, remove, edit, (unlock), change password
• Organizations - add, remove, edit, reset terms of use
• Assets - add, remove, edit, configure data display, list gateways
• Asset groups - add, remove, edit, assign assets
• Gateway - remove, edit, configure, gateway details
• Asset categories - add, remove, edit and configure asset categories

Not all sections are available for every user. Administrative sections are only available to ADMIN and TECH_ADMINs.

Detailed explanations of the individual functions can be found in the 'Management elements' chapter

Via the flag symbol to the right of the logout button, the language can be selected from a choice of English, German, Italian, French, Korean, Polish, Turkish and Chinese.

In the Support section you can find contact information for support (contact) and documentation. In addition, you can view the current software version via the information icon .

The bell icon leads to the notification center. Alarms and notifications are displayed to the user here.
notifications are displayed to the user. The number of notifications is shown directly via the appearing number above the bell symbol. In the notification center, the user can, for example, inquire about current malfunctions without having to check his e-mails.

After the user has clicked on the notification, it is automatically marked as read. Alternatively, the checkbox to the right of the message can be used to mark it as read and thus delete it.

How to activate the notifications and alarms can be found in the Trigger action chapter

05.4 My account

Via Edit profile you will be redirected to the profile page of your account. Here you can change your account password and view the organization specific values. Under Organization Specific Values, a specific value can be created by the admin with a time start and end range. This value can be calculated and output with other values and signals in the portal.

05.5 User Manual Chatbot - Technical Documentation

Date: 23.04.2026
Author: Abhishek Byrareddy

---

1. Introduction

The SCITIS User Manual Chatbot is an AI-powered assistant designed to help users interact with SCITIS documentation in a conversational way.

Currently, SCITIS documentation exists in static formats such as PDF manuals and web portals. Users are required to manually search through multiple documents to find relevant information. This leads to increased effort, slower onboarding, and higher dependency on support engineers.

This project transforms static documentation into an interactive chatbot. It allows users to ask questions in natural language and receive accurate answers derived directly from official SCITIS manuals.

The chatbot is designed to assist users, not replace human support.

---

2. Solution Overview

Instead of relying on general AI knowledge, the system:

1. Searches documentation
2. Retrieves relevant sections
3. Generates answers based only on retrieved content

This ensures:

• High accuracy
• No hallucinated responses
• Full traceability

---

3. System Architecture

The system follows a cloud-based chatbot architecture. The user query is sent to the Cloud Run service. The service handles the request and invokes the ADK agent. The ADK agent retrieves relevant context from the Vertex AI RAG corpus and uses the Gemini model to generate the final answer.

flowchart LR
    User[User] -->|Query| CloudRun[Cloud Run Service]
    CloudRun -->|Request Handling| Agent[ADK Agent]
    Agent -->|Retrieval| RAG[Vertex AI RAG Corpus]
    RAG -->|Relevant Context| Agent
    Agent -->|Prompt + Context| Gemini[Gemini Model]
    Gemini -->|Generated Answer| Agent
    Agent -->|Response| CloudRun
    CloudRun -->|Final Answer| User

---

4. Architecture Evolution

Initial Implementation (On-Premise)

• FAISS vector database stored locally
• Manual chunking and indexing of documents
• Custom retriever logic using MultiFaissRetriever

Updated Implementation (Cloud-Based)

• Retrieval handled using Google Vertex AI RAG corpus
• Documents stored and indexed in Google Cloud
• ADK agent uses Vertex AI retrieval tool
• Gemini model generates answers based on retrieved content

Key Change

FAISS (Local) -> Vertex AI RAG Corpus (Cloud)

---

5. Data Sources

The chatbot is built on a curated set of official SCITIS documentation sources. These input files are processed and indexed into the Vertex AI RAG corpus for retrieval.

The primary input sources include:

• CloudPlug Light Manual (PDF)
• CloudPlug Edge Manual (PDF)
• SCITIS Portal Documentation (Web - English and German)
• Mawera / PYROT Technical Manuals (PDF)

5.1 File Processing

The input documents are handled as follows:

• PDF manuals are uploaded directly to the Vertex AI RAG corpus
• Web documentation is indexed from SCITIS portal pages
• Each document is automatically parsed to extract raw text
• Each document is automatically split into semantically meaningful chunks
• Each document is automatically converted into vector embeddings
• Each document is automatically stored and indexed in the Vertex AI RAG corpus

---

6. LLM and Data Processing

The chatbot uses Google Vertex AI Gemini (gemini-2.5-flash) as the Large Language Model (LLM) for response generation.

The LLM is not used as a standalone general knowledge chatbot. Instead, it receives relevant documentation context retrieved from the Vertex AI RAG corpus and generates answers based only on that retrieved content.

6.1 LLM Usage

The process is:

• The user submits a question
• The ADK agent receives the query
• Vertex AI RAG retrieves relevant document chunks from the indexed documentation corpus
• The retrieved chunks are passed as context to Gemini
• Gemini generates a natural-language answer based only on the retrieved documentation
• If the retrieved documentation does not contain the answer, the chatbot should state that clearly

The LLM is configured to:

• Avoid answering from general knowledge
• Use only retrieved documentation context
• Generate concise and user-friendly responses
• Include source references when available

6.2 Input File Import

The input files are imported into the Vertex AI RAG corpus, where they are processed and indexed for semantic retrieval.

The input sources include:

• CloudPlug Light Manual: PDF manual containing device-related user and configuration information
• CloudPlug Edge Manual: PDF manual containing hardware, setup, and integration information
• SCITIS Portal Documentation: Web-based documentation available in English and German
• Mawera / PYROT Technical Manuals: PDF manuals containing technical system and operational information

6.3 Data Import Process

The import process works as follows:

• Official SCITIS manuals and documentation files are collected
• PDF manuals are uploaded directly into the Vertex AI RAG corpus
• Web documentation is indexed from the SCITIS portal documentation pages
• Vertex AI RAG automatically parses the files and extracts text
• The extracted content is split into semantic chunks
• The chunks are converted into vector embeddings
• The indexed content is stored in the cloud-based RAG corpus
• During a user query, Vertex AI RAG searches this corpus and returns the most relevant chunks to the chatbot

---

7. Deployment

The chatbot is deployed as a cloud-native application using Google Cloud Run. Cloud Run hosts the chatbot backend as a containerized service and provides a service URL that can be used for testing or future integration with the SCITIS portal.

7.1 Deployment Architecture

The application is packaged into a Docker container. The container includes the FastAPI backend, ADK agent configuration, and required Python dependencies.

The deployment uses the following Google Cloud services:

• Cloud Build to build the container image
• Artifact Registry to store the Docker image
• Cloud Run to host and run the chatbot service
• Vertex AI RAG Corpus to retrieve relevant documentation
• Gemini to generate the final response

7.2 Deployment Flow

The deployment process works as follows:

1. The chatbot application is containerized using Docker.
2. Cloud Build builds the container image from the project source code.
3. The image is stored in the Artifact Registry.
4. Cloud Run deploys the image as a managed service.
5. Cloud Run provides a service URL for accessing the chatbot.
6. At runtime, the service invokes the ADK agent, retrieves content from Vertex AI RAG, and generates answers using Gemini.

---

8. How the System Works

The SCITIS chatbot follows a cloud-based Retrieval-Augmented Generation (RAG) workflow to process user queries and generate accurate responses based on official documentation.

flowchart TD
    Query[User Query] --> Agent[ADK Agent]
    Agent --> Retrieval[RAG Retrieval Tool]
    Retrieval --> Docs[Relevant Documentation]
    Docs --> Agent
    Agent --> Gemini[Gemini Model]
    Gemini --> Agent
    Agent --> Answer[Final Answer to User]

---

9. Technologies Used

Area	Technology
Backend	Python, FastAPI
AI / LLM	Google Vertex AI, Gemini (gemini-2.5-flash)
Retrieval / Storage	Vertex AI RAG Corpus (managed retrieval system)
Agent Framework	Google ADK
Cloud & Deployment	Google Cloud Run, Cloud Build, Artifact Registry

---

10. Limitations

• Limited diagram understanding
• Depends on documentation availability