The Challenge
Next Sense’s building performance experts deliver highly detailed, asset-level diagnostics that enable clients to implement targeted energy efficiency and decarbonization measures. Each quarterly building analysis requires approximately 60 hours of expert work, combining structured sensor data with unstructured documentation such as system specifications, commissioning reports, heat pump manuals, and installation notes.
This depth of analysis is a key differentiator for Next Sense. However, many elements of the process involve manual synthesis of data and documentation. As demand grows across larger portfolios, scaling this expert-driven model while maintaining analytical rigor becomes increasingly challenging.
At the same time, Next Sense is expanding its technological capabilities to integrate generative AI into its workflow for the first time. This requires building new competencies in LLMOps, agent orchestration, and production-grade AI infrastructure – all while ensuring that expert oversight, data security, and analytical quality remain uncompromised.
The challenge was therefore twofold:
- Remove operational bottlenecks to scale expert building analysis across growing portfolios.
- Establish a robust internal AI foundation that strengthens, rather than replaces, expert judgment.
Our Approach
At Enjins, we build agentic AI with experts, not for them.
From day one, we work side by side with Next Sense’s building performance experts and engineering team. Every architectural and design decision reflects their real workflows and priorities.
Using our AI design sprint methodology, we map daily expert processes to identify automation opportunities. We quickly develop a first agentic prototype using real data, enabling experts to test it, provide feedback, and validate that it solves the right problems.
Leveraging the enthusiam generated by the prototype we moved on to a full agentic framework, with multiple agents deployed that support building experts in different tasks.
Key Elements of the Solution
1. Scalable Multi-Agent Framework
We design a production-ready multi-agent architecture centered around an orchestrator agent that manages user interactions and routes tasks to specialized agents.
This structure allows Next Sense to launch with a focused set of agents while creating a scalable foundation for future expansion. Our model-agnostic, tool-agnostic framework ensures the system grows safely and remains stable.
2. Self-Hosted, Secure Production Infrastructure
Data sovereignty is critical.
The entire system is deployed in a self-hosted environment on Azure Kubernetes Service, with dedicated infrastructure for LLM observability, vector storage, and automated data ingestion from internal document repositories.
This architecture guarantees privacy, compliance, and full control over sensitive building data.
3. Development and Deployment of Priority Agents
Together with the building experts, we launched three core agents:
- Orchestrator Agent; Coordinates user interactions and routes requests to the appropriate specialized agent.
- Raw Data Agent; Executes KQL queries against Next Sense’s Azure Data Explorer warehouse to analyze structured sensor data.
- Contextual Data Agent; Applies retrieval-augmented generation (RAG) to preprocessed technical documents, extracting relevant insights from unstructured sources.
Experts validate every iteration, shape priorities, and refine outputs. From day one, the system actively strengthens their analytical capabilities rather than replacing them.
4. Integrated Feedback Loops
We build continuous learning into the system.
The platform stores every prompt and output. Experts provide instant qualitative feedback through simple signals (thumbs up/down) or detailed comments. This feedback loop continuously improves prompts, architecture, and future agent development based on real-world usage.
5. Knowledge Transfer and Team Enablement
We don’t just deliver a system, we enable ownership.
Through deep-dive training sessions, we equip Next Sense’s engineering team with hands-on expertise in agent orchestration, LLMOps, and retrieval-augmented generation.
Following implementation, Next Sense’s software engineering team assumes full ownership of the system. Their engineers independently extend, maintain, and expand the agent framework, integrating new analytical capabilities as portfolio needs evolve.
This ensures that the solution becomes a long-term strategic asset embedded within Next Sense’s core technology stack – supporting sustained innovation in AI-augmented building performance optimization.
KPIs & Impact
• Accelerated expert workflows: Agentic support significantly reduces the time required to complete in-depth quarterly building analyses, enabling experts to cover more assets without compromising quality.
• Rapid deployment: From design sprint to a production-ready multi-agent system in just 12 weeks.
• Scalable architecture: Three agents live at launch, with infrastructure and feedback mechanisms in place to prioritize and deploy additional agents as needed.
KPIs & Impact
- Accelerated expert workflows: Agentic support significantly reduces the time required to complete in-depth quarterly building analyses, enabling experts to cover more assets without compromising quality.
- Rapid deployment: From design sprint to a production-ready multi-agent system in just 12 weeks.
- Scalable architecture: Three agents live at launch, with infrastructure and feedback mechanisms in place to prioritize and deploy additional agents as needed.
Through close collaboration, Next Sense now has a production-grade agentic AI system, done right the first time. A system co-designed with building experts, continuously validated through feedback, and fully transitioned to their software engineering team for long-term growth.
