[DSC DACH 24] Increasing user adoption with GenAI offerings - Martin Flechl

One of the most successful GenAI business models is increasing user productivity by enabling artefact generation via natural language prompts. This allows adoption of product features by every user - even of those which previously have only been accessible by people with the required technical skills. In this talk we will discuss how to design GenAI systems for this purpose, based on examples from the observability domain.

1. How to make $$$ with LLMs Be Nvidia Offer cloud services Enhance productivity Increase user adoption (and everyone else)

2. Increasing user adoption with GenAI Martin Flechl PhD, Priv.-Doz. Head of Generative AI & TechEvangelist, Dynatrace www.linkedin.com/in/mflechl

3. Dynatrace Marketecture

4. Hypermodal AI for Unified Observability and Security  

5. Natural Language Onboarding / Info (Q & A) Queries (DQL) Customer-specific answers* Dashboards, Workflows** * partially available ** potential future directions

6. Use case #1: General Q & A

7. Use case #2: Query generation

8. DQL : Hosts ordered by memory consumption

9. Davis CoPilot : Hosts ordered by memory consumption

10. Ingredients for successful DQL’ing DQ Language grammar (syntax) vocabulary commands functions operators data types Data … sources objects (tables) models (structure/relations/semantics) MAGIC MAGIC timeseries avg and, == string, array   metrics, logs ​   dt.entity.host , ​ log file/stream ​

11. Guardrail Prompt User Input Planning Prompt Generation Prompt DQL data objects, models, fields semantics and commands User Input User Input data objects, models, fields command and function syntax plan vector database SemDict plan vector database DQL doc & SemDict User Input Verification Planning Prompt Where is which data, and how to combine it Generation Prompt Explain DQL syntax, transform plan to DQL vector database “golden” queries examples Guardrail Prompt Is it DQL-related? Davis CoPilot DQL Generation Workflow

12. Generation Prompt User Input DQL User Input Verification DynaM *** LLM CONFIDENTIAL

13. Generate a query based on the domain specific language Dynatrace Query Language (DQL). Start with problem “P-230771722” and find the related application based on the affected host. Then, correlate RUM error events based on application showing the distinct number of user sessions. Kubernetes node 77652 depends on loyalty service …. How many users were impacted by yesterday’s outage? Generated Query Davis Smartscape Extraction of affected users Affected users: 486 Question Davis CoPilot-enriched prompt Answer

14. Use case #3: Customer-specific Q & A

15. User Copilot Storage Search LLM ask question verify retrieve summarize load store search completion completion read answer get history save history condense completion Data query enrich RAG*! * RAG == Retrieval-Augmented Generation Davis CoPilot Conversations Workflow

16. User Agent Storage (memory, knowledge) Tools Planning Reasoning request response, action SQL/NLP query response API/NLP query response problem task list

17. Use case: dashboard generation Create a dashboard for critical Black Friday services

18. behind the scenes… Create a dashboard for critical Black Friday services Understanding input and defining queries (generative ai) Retrieve user sessions and conversion information (causal ai) Finding relevant services for conversion path (causal ai) Understand current behavior, SLOs and past problem impact (predictive ai) Predict behavior for high - load scenarios (predictive ai) Create markup code for dashboard (generative ai)

19. How to make $$$ with LLMs Be Nvidia Offer cloud services Enhance productivity Increase user adoption &