Add OJT support agent documentation

This commit is contained in:
Gemini Bot
2025-11-21 11:25:22 +01:00
parent 964cda2be8
commit 1b9208db69

View File

@ -0,0 +1,222 @@
# OJT Support Agent Documentation
## Overview
The OJT (On-the-Job Training) Support Agent is a workflow-based AI application designed to provide FAQ support for both new hires and ongoing employee training. The app uses knowledge retrieval and language models to deliver relevant answers based on user queries and support type selection.
## App Configuration
### Basic Information
- **Name**: FAQ
- **Mode**: Workflow
- **Version**: 0.5.0
- **Icon**: 🤖 (orange background #FFEAD5)
### Dependencies
- **LLM Provider**: LangGenius Gemini (version 0.5.12)
- **Models Used**:
- Gemini 2.5 Flash (for knowledge retrieval)
- Gemini 2.5 Pro (for response generation)
## Workflow Structure
### Input Fields
The application accepts the following user inputs:
1. **support_type** (Required)
- Type: Select dropdown
- Options:
- "New hire Training"
- "On-the-Job Training (OJT)"
- Max length: 48 characters
2. **Question** (Required)
- Type: Paragraph text input
- Max length: 400 characters
- User's specific question or query
3. **process_related_FAQs** (Optional)
- Type: File upload
- Allowed formats: CSV files only
- Upload methods: Local file or remote URL
- Max files: 2
### Workflow Nodes
#### 1. Start Node (User Input)
- **ID**: 1763464241668
- **Purpose**: Collect user inputs
- **Position**: Entry point of the workflow
#### 2. Knowledge Retrieval Node
- **ID**: 1763464408773
- **Purpose**: Search knowledge base for relevant FAQs
- **Configuration**:
- Dataset ID: HhYDV55j3H4hoASbF6jsL+Bh+L3nSYCjs2yq8FqZBmzgNiViWkI9TuOzd5pWfApA
- Retrieval mode: Multiple
- Top K results: 10
- Score threshold: Not configured
- Reranking: Disabled
- Query variables: support_type from user input
- Temperature: 0.7
#### 3. LLM Node
- **ID**: 1763466795085
- **Purpose**: Generate responses based on retrieved knowledge
- **Model**: Gemini 2.5 Pro
- **Temperature**: 0.7
- **Prompt Template**:
```
System: You are a helpful assistant
User: show only those rows where you can find in row {{support_type}} and is related to {{Question}}
output formatting: provide only questions and answers (in each paragraph)
context you can find here: {{context}}
User: summarize also content from files {{sys.files}} take consideration about {{support_type}} and {{Question}}
```
#### 4. End Node (Output)
- **ID**: 1763464894998
- **Purpose**: Return final response to user
- **Output variable**: text from LLM node
## Features
### Enabled Features
- **Knowledge Retrieval**: Enabled for FAQ database access
- **File Upload**: Limited to CSV files for process-related FAQs
### Disabled Features
- **Speech to Text**: Not enabled
- **Text to Speech**: Not enabled
- **Suggested Questions**: Not configured
- **Suggested Questions After Answer**: Not enabled
- **Sensitive Word Avoidance**: Not enabled
- **Image Upload**: Not enabled
## File Upload Configuration
### Allowed File Types
- **Extensions**: .CSV only
- **Types**: Custom
- **Upload Methods**: Local file, Remote URL
- **File Size Limit**: 15MB
- **Batch Count Limit**: 5 files
- **Number Limits**: 3 files
### Upload Restrictions
- Audio files: 50MB limit
- Image files: 10MB limit
- Video files: 100MB limit
- Workflow file upload limit: 10
## Usage Instructions
### For End Users
1. **Select Support Type**: Choose between "New hire Training" or "On-the-Job Training (OJT)"
2. **Enter Question**: Type your specific question (max 400 characters)
3. **Upload CSV (Optional)**: Upload relevant FAQ files if available
4. **Submit**: Receive AI-generated answers based on knowledge base
### For Administrators
#### Knowledge Base Management
- Update the knowledge base dataset (ID: HhYDV55j3H4hoASbF6jsL+Bh+L3nSYCjs2yq8FqZBmzgNiViWkI9TuOzd5pWfApA)
- Ensure FAQs are properly tagged with support_type categories
- Maintain CSV format for uploaded files
#### Performance Optimization
- Monitor retrieval effectiveness (Top K: 10)
- Adjust temperature settings if needed (currently 0.7)
- Consider enabling reranking for better result accuracy
## Integration Points
### Dataset Integration
- Primary knowledge base connected via dataset ID
- Supports metadata filtering based on support_type
- Multiple retrieval mode for comprehensive results
### File Processing
- CSV file content is summarized and incorporated into responses
- File content is filtered by support_type and question relevance
- Supports both local and remote file sources
## Security Considerations
- No sensitive word avoidance enabled
- File upload restricted to CSV format only
- User inputs have length restrictions to prevent abuse
- Knowledge retrieval limited to configured dataset
## Limitations
1. **File Format**: Only CSV files accepted for upload
2. **Input Length**: Questions limited to 400 characters
3. **Language**: Configuration suggests English-only support
4. **Real-time Updates**: Knowledge base requires manual updates
5. **Multimedia**: No image, audio, or video support
## Troubleshooting
### Common Issues
1. **No Results Found**
- Check if support_type is correctly selected
- Verify knowledge base contains relevant FAQs
- Consider adjusting Top K retrieval count
2. **File Upload Errors**
- Ensure file is in CSV format
- Check file size doesn't exceed 15MB
- Verify file contains relevant support_type data
3. **Poor Response Quality**
- Review prompt template for clarity
- Consider adjusting temperature settings
- Update knowledge base with more comprehensive FAQs
### Performance Issues
- Monitor response times from LLM nodes
- Check knowledge base retrieval efficiency
- Consider enabling reranking for better result filtering
## Future Enhancements
### Recommended Improvements
1. **Enable Suggested Questions**: Improve user experience with question suggestions
2. **Add Multilingual Support**: Expand to multiple languages
3. **Enhanced File Support**: Allow additional file formats (PDF, DOCX)
4. **Real-time Knowledge Updates**: Implement automatic knowledge base updates
5. **Analytics Dashboard**: Track usage patterns and popular questions
### Advanced Features
1. **Conversation Memory**: Enable multi-turn conversations
2. **User Personalization**: Tailor responses based on user role
3. **Integration APIs**: Connect with HR systems for real-time data
4. **Voice Support**: Enable speech-to-text and text-to-speech
5. **Advanced Analytics**: Implement sentiment analysis and feedback collection
## Support and Maintenance
### Regular Maintenance Tasks
- Update knowledge base with new FAQs
- Monitor model performance and accuracy
- Review and update prompt templates
- Check file upload functionality
### Contact Information
- For technical issues: Contact system administrator
- For knowledge base updates: Contact content management team
- For user support: Contact help desk
---
*Last Updated: November 2025*
*Version: 1.0*