Problem space
- Limited system visibility
- Manual troubleshooting
- Fragmented operational knowledge
- Specialist dependency
- Difficulty connecting data and context
Work
- Identified operator problems
- Translated needs into platform capabilities
- Considered data availability, technical feasibility and operational value
- Designed a planner-based multi-agent architecture
- Designed six specialised agents
- Data retrieval, tool use and grounded answer generation
Technology
- LangGraph
- LangChain
- LangSmith
- OpenAI APIs
- Anthropic APIs
- GraphRAG
- Knowledge Graphs
- Time-series data
- Ontology-based context
Product perspective
Explored how agentic AI could become a reusable capability inside a smart-building / energy platform rather than merely a standalone chatbot.
Reflection
Related to Talking SCADA in problem space, but a distinct concept with its own architecture and product framing.