Daniel Dayto
Seven years building production platforms — for the world's largest retailers, global airlines, and service providers.
Software engineer & technical co-founder with seven years of experience owning products end to end—from customer discovery and architecture through deployment and production operations. I’ve built computer-vision systems used by major retailers, travel platforms for global airlines, and release infrastructure supporting 50+ applications. Most recently, I architected Playmaker, an AI voice platform that converts inbound customer conversations into booked revenue.
Production results
Voice agents I built and operate in production at Berrett Home Services, a Texas HVAC and plumbing company — answering their inbound line, qualifying callers against the company's own rules, and booking appointments directly into ServiceTitan.
- Calls handled1,152Customer conversations handled autonomously.
- Appointments booked158Appointments booked directly into the client CRM system
- Jobs dispatched139Booked jobs assigned and ready for service
- Additional revenue generated$56K+Verified revenue from AI-booked jobs
Outcome attribution
From conversation to recorded revenue
- Inbound call
- AI handles conversation
- Appointment booked
- Job dispatched
- Revenue recorded
Berrett Home Services · Mar 2026–present
How attribution works
- Each result is traced through production events from the initial call through the ServiceTitan job record.
- $56K+ is a measured floor. It reflects recorded value for 51 of 158 AI-booked jobs currently synced from ServiceTitan. Remaining jobs are excluded until their value is available.
- Attribution indicates that the AI booked the job; it does not assume all recorded revenue is net-new.
How I work
One continuous loop, from customer conversation to measured outcome.
This loop is usually split across a consultant, an architect, and a delivery team. I run all of it myself: I talk to the customer, map the workflow, design the architecture, write the software, deploy it, and measure the result.
- 01Discover
Understand the operation before the technology
I start with the people closest to the problem — owners, CSRs, dispatchers — and watch how the work actually gets done. At Berrett, the rules that mattered most (urgency tiers, service-area boundaries, membership handling, when to escalate) existed only in people's heads. The operation already running the business is the spec; most of it is written down nowhere.
The discovery work → - 02Model
Make the workflow explicit — including how it fails
The ask was “an AI phone agent.” The real problem was inbound demand lost whenever staff couldn't answer. I mapped the workflow a human follows — qualification, account lookup, service-area validation, scheduling, booking, dispatch, escalation — along with its exceptions and failure modes, and separated root causes from symptoms. That map decided what to build; the technology didn't.
- 03Design
Architecture follows the operation
From the map I draw boundaries: what gets automated and what stays human-controlled, where deterministic rules apply and where AI judgment is acceptable, which system remains the source of record, how failures recover, and what will be measured. For Playmaker, that translated into conversational AI over Python services, ServiceTitan integration, booking orchestration with internal persistence and async recovery, and explicit human-escalation paths.
The architecture → - 04Deploy
Ship for the environment that exists, not the diagram
Production is messy: APIs time out mid-call, CRM data is inconsistent, callers are unpredictable, AI is probabilistic. So bookings persist to our own database before any CRM write and recover asynchronously when a vendor fails. And new automation earns authority in stages — baseline, observe, compare, validate, expand. The transfer policy classified live calls in shadow mode against human decisions before it was allowed to act; AI call scoring was calibrated against human review the same way.
The transfer policy → - 05Measure
Connect system behavior to business outcomes
Deployment isn't the finish line. Every call resolves to a recorded event, so outcomes trace end to end — call, AI conversation, appointment, dispatched job, recorded revenue — and the questions that matter get answered with data: did completion rise, did conversion improve, how much workload came off the team, what revenue is attributable to the system.
Production results → - 06Iterate
Production evidence drives the next cycle
Real calls surface edge cases no discovery session predicts. Failures become permanent regression scenarios — a change that can't beat the suite doesn't ship — and metrics point to the next bottleneck. I take what production shows back to the customer and go around again.
The postmortem →
Selected work
Systems running in live operations — voice agents on real phone lines, a copilot inside live call workflows, and the platform behind 50+ customer apps.
Playmaker · Co-founder & Technical Lead
AI Voice Agents for Home-Service Call Handling
- Context
- Home-service call operations losing after-hours and overflow demand
- Solution
- AI voice agents integrated with ServiceTitan, scheduling, routing, and human escalation
- Result
- 1,152 calls handled, 158 appointments booked, and 139 jobs dispatched to technicians
Playmaker · Co-founder & Technical Lead
RAG Copilot for CSR & Sales Teams
Reps lacked consistent access to pricing, membership, and booking rules mid-call. A retrieval copilot embedded in the booking workflow took the client's inbound close rate from 72% to 92%.
03WOLF · Lead Mobile Developer
Multi-Tenant Deployment Infrastructure for 50+ Customer Apps
Releases depended on manual, engineer-specific steps. A standardized multi-tenant architecture and deterministic delivery pipeline lifted release velocity 35% across 50+ branded apps.
04Emirates Skywards · Senior Full-Stack Developer
High-Scale Travel Booking Platform
Rail booking had to work inside a live airline loyalty platform. SSO identity, multi-provider rail APIs, and distributed booking orchestration shipped under enterprise review — supporting millions of monthly searches.
Writing
Field notes from building and operating these systems — postmortems, architecture decisions, and what production traffic actually teaches.
AI Evaluation & Reliability
Building a Regression Testing Pipeline for Production AI Agents
An engineering account of adding regression testing to a production voice agent: why prompt changes regress unrelated behavior, the 0–100 call-scoring model, deterministic checks vs. LLM evaluation, critical-failure caps, the CI release gate, and calibrating the evaluator against human review.
· 12 min read
What I specialize in
The useful unit of AI is a completed business workflow — not a model response. I work these problems end to end, on the engineering foundation below.
Capture & convert demand
- AI voice agents
- Lead qualification
- Appointment booking
- Sales follow-up
- 24/7 inbound coverage
Automate manual workflows
- Agent orchestration
- Workflow automation
- Human-in-the-loop
- Exception & failure handling
- Conversational AI
Augment employees with AI
- CSR & employee copilots
- RAG
- Knowledge retrieval
- Live-call assistance
Integrate into operational systems
- ServiceTitan
- FieldRoutes
- Twilio
- CRM & dispatch
- REST APIs & webhooks
- Multi-tenant configuration
Measure & optimize outcomes
- Evaluation & QA
- Observability
- Revenue attribution
- Regression detection
- Operational dashboards
Engineering foundation
- Python
- FastAPI
- Node.js
- TypeScript / React / Next.js
- PostgreSQL
- Redis
- AWS
- Docker
- GitHub Actions · CI/CD
About
Turning messy operations into systems that hold up
Robotics data tooling, enterprise travel integrations, a 50-app multi-tenant platform, and now production AI for service businesses — the same work each time: understand a real operation well enough to build the system that runs it. Co-founder and technical lead at Playmaker.
Contact
Deploying AI into a real operation?
Hiring for a senior role
Senior AI, Forward Deployed Engineering, product engineering, platform, and technical leadership roles.
Need a problem solved
Selected contracting and advisory work involving production AI deployments, integration builds, platform architecture and audits, and technically difficult product initiatives.
Run a home-service business and want AI answering your phones? That's Playmaker.