SmartRAG Enhacements

The issue aims to track the following enhancements or improvements made to smartRAG over the course of time.

Tasks

• Interactive visualization for Cluster embeddings and Specific QA points
• Formulation of Questions on Public adminstration document ( Open-data Handbuch )
• Opengpt-X-Teuken-Rag and Opengpt-X-Teuken-Instruct with RAG pipline:
  • Results are documents in word document
  • Obtained Prompt template from Dreseden team
• Deployment of new Opengpt-X-Teueken commercial and research model on DGX server
• Onboarding Comercial and Research models into AI-Builder
• SmartRAG republish and Rag-node delimeter issue
• RAG output speed improvisation
• Generate English responses from SmartRAG Teuken models:
  • Published separate models with simple prompt and EN tag on AI-Builder dev and DeployAI-Prod
  • Prompt Engineering for V0.4 models and hyperparameter tuning ( seed )
  • Prompt Engineering for V0.6 sigma model
  • Preparation for English webinar with Schengen Visa document
  • V0.6 and V0.4 EN models and pipeline published on AI-Builder dev and AI-LAB NRW
  • Documentation of Results
• Advance Prompt tuning feature:
  • Frontend webUI and backend scripting on rag node
  • Backend scripting on LLM's rag_chain node#
• Generate German response for SmartRAG Teuken V0.6 Sigma :
  • Prompt Engineering for V0.6 model with seed
  • Published model on AI-LAB NRW and AI-Builder dev
• Unifying rag_chain scripts to work with common LLMs ( uses langchain ) and opengptx llms.
  • Langchain or Prompting techniques?
• Adding more document feature
• parameter tunning feature
  • temp, topk, seed, ?

Innovation LAB 2025

• Collect prompt in execution run data
  • Prompt is appended to the rating.json file
• Efficient metric display and collection - Userfeedback node
  • WebUI changes and metric completion method ( display ratings after evaluation metrics --> indicate users to complete for overall efficiency)
  • get_metadata_metric pinged frequently to log current metrics
  • AI-Runner backend modification for capturing latest metrics
• Does saving work?