Who they are
Dwelly is a UK-based AI-native real estate company building the operating system for residential lettings. The company acquires independent letting agencies and integrates them into an AI-powered platform that automates tenant communications, property management, maintenance and other operational workflows.
As Client scaled, they needed to automate a growing volume of inbound customer calls without replacing their existing telephony infrastructure. That required an engineer who could bridge AI, voice technology and their existing backend systems.
What was on the table
“We want to add an AI voice agent” sounds simple. In production, it’s three engineering problems stacked.
First, integrating an LLM-driven voice agent (ElevenLabs) into a working telephony stack (VoxImplant) without breaking existing call flows. Second, making the agent useful: recognizing the caller by phone number, pulling context from the internal server, and handing off cleanly to a human when needed. Third, logging everything (transcripts, audio recordings, outcomes) for compliance and quality review.
Most senior backend devs can do one of those. Few can do all three on a production timeline.
Why they chose UASOFTDEV
We had a fullstack engineer with prior experience in both VoxImplant telephony APIs and LLM-based agents (ElevenLabs + OpenAI). Specific stack overlap, uncommon profile.
Alexander (CTO) reviewed the architecture proposal before kickoff. First match within 60 hours.
What the engineer delivered
- Designed and built the integration layer between ElevenLabs voice agents and VoxImplant telephony, handling call routing, agent invocation, and fallback to human operators
- Built caller identification logic: incoming call number cross-referenced against the client’s internal database, with context (order status, customer history) passed into the agent’s session
- Implemented multilingual IVR flows – “press 1 for orders” replaced with natural-language conversation
- Set up end-to-end logging: transcripts, audio recordings, conversation outcomes saved to the client’s admin server via API calls, queryable for QA and compliance
- Built handoff protocols: when the agent detects edge cases, complaints, or specific keywords, the call is routed to a human operator with the conversation context already loaded
The outcome
After 3 months of buildout and tuning, the AI voice agent went live and now handles ~70% of inbound first-level calls — order confirmations, status inquiries, basic troubleshooting.
The engagement transitioned from build phase into ongoing maintenance and expansion – adding new agent capabilities, training on new use cases, expanding to outbound (order confirmation calls).
By the numbers
Previous vendors quoted 6-month timelines and didn’t know the stack - UASoftDev came in with ready expertise and shipped to production in 12 weeks.
Running a voice-driven workflow? Add AI without rebuilding your stack.
If you’re running a customer-support operation, telephony service, or any voice-driven workflow and want AI to handle the routine layer - this is what we run through Staff Augmentation. Compatible telephony stacks: VoxImplant, Twilio, Vonage. Compatible AI providers: ElevenLabs, OpenAI, Anthropic.