Build a Voice AI System for Your Maintenance Team
Syntora is a consultancy that builds voice AI systems for property maintenance operations. We build systems that answer tenant calls, transcribe issues, and create detailed work orders automatically.
We built a voice agent for a 25-person property management firm handling 800 units. Their 3-person maintenance team was spending 2 hours daily on call-backs. The system went live in 3 weeks, reduced manual call handling by 90%, and now triages over 50 inbound maintenance requests per day.
The build's complexity depends on your existing property management software (PMS) and call volume. Integrating with a modern PMS like AppFolio via its API is a 2-week project. Connecting to older, on-premise systems without APIs requires more custom scripting and can take up to 4 weeks.
What Problem Does This Solve?
Most property managers first try virtual receptionist services like Ruby. These services are staffed by humans who take messages and send emails, but they cannot create a structured work order in your PMS. A coordinator still has to manually re-enter the data. A single misheard address, like '13th Street' instead of '30th Street', wastes two hours of a technician's time and fuel.
Others attempt to build a simple IVR with a tool like Twilio Studio. The 'Press 1 for plumbing, Press 2 for electrical' flow is too rigid for real-world issues. It cannot understand a frantic tenant reporting that 'water is coming through my ceiling from the apartment above'. This urgent request gets routed to a generic voicemail, delaying the response by hours and leading to thousands in water damage.
The fundamental problem is that these tools cannot ask intelligent, clarifying questions. When a tenant reports a 'broken appliance', the system needs to ask 'Which appliance?' and 'What is it doing?'. Without this conversational ability, work orders are incomplete, forcing your team to make a follow-up call anyway and delaying repairs by a full day.
How Does It Work?
We provision a dedicated phone number through Twilio. When a tenant calls, an AWS Lambda function receives the call and uses a real-time transcription service to capture the conversation. The initial transcription and issue classification, such as plumbing or HVAC, happens in under 3 seconds. We use Python with the FastAPI framework to manage the webhook from Twilio.
The live transcript is sent to the Claude API with a detailed prompt. This prompt instructs the AI to act as a maintenance dispatcher: extract the unit number, tenant name, contact info, issue category, and a detailed description. The AI is prompted to ask clarifying questions if details are missing, for instance, 'Can you confirm the address and unit number for me?'. The entire AI-driven conversation takes an average of 90 seconds.
Once all data is collected, a separate Python function formats it into a JSON payload and posts it directly to your PMS API, whether it is AppFolio, Buildium, or a custom system. We write a record of every call, transcript, and generated work order to a Supabase database for logging and audit trails. A successful work order is created within 500ms of the call ending.
Finally, the system sends an immediate confirmation SMS to the tenant and a Slack notification to the maintenance coordinator with a link to the new work order. We use `structlog` for structured JSON logging. This allows us to set up CloudWatch alerts that trigger if the API error rate exceeds 1% or if Claude API latency goes above 4 seconds. Monthly AWS costs for processing 1,500 calls are typically under $75.
What Are the Key Benefits?
Your AI Agent is Live in 3 Weeks
From kickoff to handling live tenant calls in 15 business days. The system immediately reduces your team's phone time and manual data entry.
Pay Once, Own the Code Forever
A fixed-price project with no per-call or per-seat fees. You receive the full Python source code in your company's GitHub repository.
Never Miss an After-Hours Emergency
The system operates 24/7, correctly identifying and escalating urgent issues like major leaks, even at 2 AM on a Sunday.
Integrates Directly With Your PMS
We build custom API integrations for AppFolio, Buildium, Yardi, and other property management systems. No more copy-pasting from emails.
Full Audit Trail of Every Call
Every call, transcript, and AI decision is logged in a Supabase database. You have a searchable record to resolve any tenant disputes.
What Does the Process Look Like?
Discovery and PMS Access (Week 1)
We map your current maintenance intake process. You provide API credentials for your Property Management Software and we define the work order fields.
Core AI and Logic Build (Week 2)
We build the core FastAPI application and Claude API prompts. You receive a private phone number to test the system with sample maintenance requests.
Integration and Deployment (Week 3)
We connect the voice agent to your live PMS, deploy it on AWS Lambda, and route your maintenance line to the new number. Your team gets a live demo.
Monitoring and Handoff (Weeks 4-8)
We monitor every call for 30 days to tune AI performance. You receive a final runbook with system architecture diagrams and support instructions.
Frequently Asked Questions
- How much does a custom voice AI system cost?
- Pricing is a fixed, one-time fee based on complexity. A system that only transcribes and emails is simpler than one that creates tiered work orders in a legacy PMS via a custom integration. We provide a detailed quote after a 30-minute discovery call where we review your current workflow and software stack.
- What happens if the AI misunderstands a tenant?
- If the AI cannot confidently extract the required information after two attempts, the system automatically routes the call to a human coordinator. A Slack alert is sent with the call transcript, so your team can intervene immediately. This fail-safe ensures no request is ever lost, maintaining a 99.5% successful intake rate.
- How is this different from a service like Talkdesk or a generic call center AI?
- Talkdesk provides a platform, but you still need to build the specific property management logic. They are designed for large sales or support teams. Syntora builds a purpose-fit agent that understands terms like 'HVAC' and 'leaky faucet' and integrates directly into your PMS, creating structured work orders, not just support tickets.
- Can it handle different languages or heavy accents?
- Yes. The underlying transcription and language models support dozens of languages. We can configure the system to detect the caller's language and respond appropriately. While transcription is not perfect for very heavy accents, the AI is trained to ask for clarification, such as spelling out a street name, to ensure accuracy.
- What are the ongoing costs after the build?
- You pay for the underlying cloud services directly. A typical system handling 1,500 calls per month costs under $75 in AWS Lambda, Twilio, and Claude API fees. We offer an optional flat monthly maintenance plan that covers monitoring, bug fixes, and prompt tuning after the initial 30-day support period.
- Do we need an engineer on staff to run this?
- No. The system is deployed on serverless infrastructure (AWS Lambda) that requires no server management. It's designed to be self-sufficient. You receive a runbook that explains the architecture, but you do not need to interact with it unless you want to make changes to the core logic. We handle all deployment and monitoring.
Related Solutions
Ready to Automate Your Small Business Operations?
Book a call to discuss how we can implement ai automation for your small business business.
Book a Call