Operio Agent

Inspiration

Commercial mall operations still run on fragmented workflows. When a tenant reports an issue, managers often have to read lease clauses manually, check equipment documentation, verify technician availability, and decide whether the work can be dispatched immediately or needs landlord approval. We wanted to build an agent that could do that reasoning loop end to end instead of acting like a simple chatbot.

What it does

Operio Agent is a lease-aware AI operations agent for commercial malls.

A tenant can report an incident in natural language, and the agent will:

• retrieve the relevant lease clauses
• pull equipment manual context when diagnostics are needed
• check technician availability and certifications
• create or update work orders
• escalate landlord-liable or high-cost issues into a human approval flow
• trace the full reasoning path in Arize Phoenix

The result is a system that does more than answer questions. It turns operational context into traceable dispatch decisions.

How we built it

We built Operio Agent as a polyglot system with:

• a FastAPI backend orchestrating the agent workflow
• Gemini for reasoning
• MongoDB for operational data like tenants, staff, sessions, leases, manuals, and work orders
• MongoDB Atlas Search and vector search for retrieval across lease clauses and equipment manuals
• MCP servers for transactional and retrieval operations
• a Vite + TypeScript frontend for the operator experience
• OpenTelemetry and Arize Phoenix for trace-level observability and evaluation visibility

We also created a shared scenario benchmark so the same operational cases can be replayed and evaluated consistently.

Challenges we ran into

The hardest part was making the agent behave like an operations coordinator instead of a generic assistant. It had to reason across liability rules, approval thresholds, duplicate incidents, technician qualifications, and retrieval evidence while staying explainable. Another challenge was structuring the traces and evaluation signals so judges can inspect not just the final answer, but the path the agent took to get there.

Accomplishments that we're proud of

We’re proud that Operio Agent can:

• use real operational context before acting
• distinguish between tenant-liable, landlord-liable, and ambiguous incidents
• route cases into the correct workflow instead of forcing every request into a work order
• keep the reasoning path visible through Arize traces
• combine cloud AI, retrieval, workflow state, and human approval into one coherent system

What we learned

We learned that useful agents need strong workflow boundaries, grounded retrieval, and clear observability. In operations software, the important question is not just “did the model answer?” but “did it take the correct action, with the correct evidence, and can someone audit that decision later?”

What's next for Operio Agent

Next, we want to expand Operio Agent into a broader property operations platform with deeper integrations into property management systems, richer approval policies, stronger evaluation loops, and production-ready multi-site deployment.

Built With

• arize-phoenix
• docker
• fastapi
• gemini
• google-cloud
• mcp
• mongodb
• mongodb-atlas-search
• opentelemetry
• python
• typescript
• vector-search
• vite