Add an AI voice agent to the phone line you already have.
Senior engineers who connect a voice agent to your existing telephony and internal systems. It recognises the caller, answers from their real data, hands complex calls to a person with context, and logs every conversation.
Results for Dwelly, a UK AI-native lettings company.
- 00:00Incoming call from +44 7••• ••• 214
- 00:01Caller recognised: tenant at 14 Harbour Road, open repair ticket
- AgentHi Sarah, are you calling about the boiler repair at Harbour Road?
- CallerYes. When is someone coming?
- 00:08Found in property system: engineer booked Thu 10:00–12:00
- AgentAn engineer is booked for Thursday between ten and twelve. Shall I text you a reminder?
- CallerPlease. And I want to complain about the last visit.
- 00:19Complaint detected: handing to a person with the conversation loaded
- AgentI'm putting you through to Tom now. He'll see everything we've discussed.
- 00:22Call logged: transcript and audio saved to your system
“Add an AI voice agent” is three engineering problems stacked.
Fitting into the phone line you have
The agent has to sit inside your existing telephony without breaking the call flows your customers already use. Rebuilding the phone system is rarely an option.
Knowing who is calling
A useful agent recognises the caller by number, pulls their record from your system and answers about their repair, booking or payment, not in generalities.
Knowing when to step aside, and keeping a record
Complaints and unusual requests go to a person with the conversation already loaded. Every transcript, recording and outcome lands in your own system for quality review and compliance.
Most senior developers can handle one of these. The engineer we placed at Dwelly handled all three.
When an off-the-shelf voice bot is enough, and when it isn't
Voice platforms such as Vapi, Retell or ElevenLabs Agents are good products. The question is whether your calls can be answered without the data inside your own systems.
| A platform on its own | A platform plus our engineer | |
|---|---|---|
| Good for | FAQs, simple bookings, few integrations | Calls that depend on data in your own systems |
| Telephony | Usually their numbers or a standard connector | Your existing telephony and call flows |
| Caller context | Limited to what the platform can reach | Your internal records, looked up on every call |
| Handoff to a person | Generic transfer | Your rules, with the conversation loaded for the operator |
| Logs | Kept in the vendor's dashboard | Transcripts, audio and outcomes in your own system |
| Who owns the logic | The vendor | You |
If your calls fit the left column, a platform on its own is the cheaper answer, and we will tell you so on the call.
How Dwelly automated about 70% of first-level calls in 12 weeks
The client
Dwelly is a UK AI-native real estate company building the operating system for residential lettings. It acquires independent letting agencies and moves them onto an AI platform that automates tenant communication, property management and maintenance.
The challenge
As Dwelly grew, inbound calls grew with it. They needed to automate them without replacing their telephony, which meant finding one engineer who could bridge voice AI, telephony and their backend at the same time.
Why they chose us
We had a senior fullstack engineer with hands-on experience in both VoxImplant telephony and LLM voice agents. Our CTO reviewed the architecture before kickoff, and the first match took 60 hours.
What the engineer built
- The integration layer between ElevenLabs voice agents and VoxImplant telephony, including call routing and fallback to human operators.
- Caller identification against Dwelly's internal database, with the caller's context passed into the agent's session.
- Multilingual, natural-language call flows in place of “press 1” menus.
- End-to-end logging of transcripts, audio and outcomes to Dwelly's admin server, ready for QA and compliance.
- Handoff rules that route complaints and edge cases to a person with the conversation already loaded.
The outcome
After three months of build and tuning, the agent went live and now handles about 70% of inbound first-level calls. The engagement continued for more than 14 months, adding new agent capabilities and outbound calls.
Previous vendors quoted six-month timelines and didn't know the stack. UASoftDev came in with ready expertise and shipped to production in 12 weeks.Read the full case
How many hours of routine calls could an agent take?
Put in your own numbers. The estimate applies the ~70% first-level share from the Dwelly case to your routine calls.
An estimate, not a quote. Your real call mix decides the number, which is what the 30-minute review is for.
Check this against my real callsFrom first call to a live agent
-
Call review, 30 minutes
Bring your top call reasons and a rough monthly volume. You'll hear which calls an agent can handle, what it takes on your current phone system, and whether an off-the-shelf tool would do without us.
-
Architecture review and match
Our CTO reviews the approach before kickoff and we send one or two engineer profiles, not a pile of CVs. At Dwelly the first match took 60 hours.
-
Build to production
The engineer joins your team and works in your repositories and telephony. At Dwelly the agent went live in 12 weeks.
-
Tune and expand
Monthly engagement with no long lock-in. New call types, languages and outbound calls follow as the agent proves itself.
Works with what you already run
- Telephony: VoxImplant in production at Dwelly; other SIP-based telephony on request.
- Voice and language models: ElevenLabs, OpenAI, Anthropic.
- Your backend: PHP, Node.js, Python, and REST or GraphQL APIs to your CRM, property or booking system.
Who this is for
- Lettings and property management: tenant repairs, viewings, rent and deposit questions.
- Home and field services: bookings, reschedules and “when is the engineer arriving”.
- Delivery-heavy e-commerce: delivery windows, order status and returns, where customers still pick up the phone.
- Other busy phone lines: clinics, logistics and service centres. Ask us.
Why teams work with us
Founders on every engagement
Oleksandr Shutalov, our CTO, reviews the architecture and answers technical blockers within two working hours. Aleksey Pavlishen, our CEO, runs the first call and owns the match.
A record, not a promise
95.65% first-match success across 23 engagements in 24 months, an average engagement of 18+ months, and 5/5 from 10 reviews on Clutch.
The code stays with you
The engineer works in your repositories and your tools. Everything they build, and the knowledge of how it works, belongs to your team.
Questions we usually get
Do we have to change our phone system?
No. The agent is connected to the telephony you already run, so existing call flows keep working. At Dwelly the agent was added on top of their VoxImplant setup.
Will callers know they're talking to an AI?
That is your policy decision. Most teams announce it at the start of the call, and we build whatever disclosure you choose.
What happens when the agent doesn't understand?
It hands the call to a person with the conversation loaded, following rules you set: keywords, complaints, specific requests or repeated failed attempts. Nobody has to repeat themselves.
Where are recordings and transcripts stored?
In your own systems, not ours. Recording disclosures and retention follow your existing compliance rules.
How long does it take?
At Dwelly it took 12 weeks from kickoff to production. Simpler call flows can be faster; the call review gives you a realistic estimate for yours.
How much does it cost?
You pay monthly for the engineer's time, plus usage fees from the voice and model providers. You get an exact rate together with the engineer profiles.
Which time zones do your engineers work in?
Ukraine is UTC+2/+3. That gives a full working-day overlap with the UK and EU and about four hours with the US East Coast, and engineers adjust to your standups.
Bring your call reasons. Leave with a plan.
In 30 minutes with our CTO you'll know which calls an agent can take, what it takes on your phone system, and whether you need us at all.
How to reach us
Website: uasoftdev.com/#contacts
Email: support@uasoftdev.com
Phone: +380 98 364 54 45
Founders on LinkedIn: Oleksandr Shutalov, CTO, and Aleksey Pavlishen, CEO
We reply within 12 hours.