+$169,320
in twelve months.
One engineer.
One pipeline.
Multi-channel real-time AI lead-intake for SMBs. Designed, built, and shipped end-to-end.
Mid-Atlantic AI's anchor customer went from $235,568 to $404,888 in twelve months — +71.9% revenue, +80% job volume, +297% Google search impressions. The lift came from infrastructure, not advertising: a multi-channel AI pipeline that captures every lead, deduplicates against the CRM, auto-creates the job, and falls back to AI callbacks when SMS hits a landline.
The receipts behind +$169K.
Channel attribution (CY 2023 → CY 2024)
| Source | 2023 jobs | 2024 jobs | Lift |
|---|---|---|---|
| 187 | 430 | +130% | |
| Friend / Family | 51 | 163 | +220% |
| Yelp | 56 | 78 | +39% |
| Search Engine (other) | 34 | 86 | +153% |
| Next Door | 24 | 74 | +208% |
| Home Depot | 1 | 8 | +700% |
Service-line growth (Aug-Aug fiscal year)
| Service | Baseline | Post | Lift |
|---|---|---|---|
| Refrigerator | $77,895 | $137,434 | +76% |
| Washing machine | $36,958 | $71,348 | +93% |
| Dryer | $29,769 | $52,933 | +78% |
| Dishwasher | $29,466 | $56,219 | +91% |
| Range | $31,913 | $43,367 | +36% |
The lift came from conversion infrastructure, not pricing or one-off jobs. Revenue and volume grew faster than average ticket — meaning more leads captured and more of them converted. Broad-based growth across every major service line rules out "one big customer" or "one viral SKU" explanations.
The system ran on the customer's existing tooling: Hibu-published website + Kickserv CRM. ChatPro deployed onto the existing site without forcing a replacement. Voice Pro and ChatPro both wrote directly into Kickserv without forcing a CRM migration. The lift came from making existing tooling more effective.
*All metrics verifiable against Kickserv revenue reports, Kickserv marketing source summaries, and Google Business Profile insights. Dated screenshots available on request.
Five channels in. One booked job out.
The pipeline behind +71.9%.
The Pro Suite isn't three products. It's one end-to-end automated lead pipeline that turns inbound contact across five channels into deduplicated Kickserv records with multi-recipient notifications, AI-extracted structured data, retry handling, and resilient landline fallback.
Watch a real call flow through it, or click any channel to see how that source feeds the same engine. The 98%+ dedup rate, the AMD-aware callback fallback, and the operator-only-shows-up workflow are what drove the revenue lift — not "voice AI sounds nice."
The stack, end to end.
Telephony
- Twilio Voice
- Twilio SMS (A2P 10DLC)
- Twilio Media Streams (raw WS)
- AMD (Answering Machine Detection)
Voice AI
- OpenAI Realtime API (gpt-4o-realtime)
- Raw WebSocket protocol
- Latency-optimized streaming
- Whisper STT
TTS
- Amazon Polly (neural voices)
- 8 languages, matched native
- Pre-cached audio pipeline
- Hash-based MinIO cache
Backend
- FastAPI (async)
- Flask (chat layer)
- MongoDB + Motor
- SQLAlchemy + APScheduler
Frontend (SitePro)
- React 19 + TypeScript
- MUI v6 + TanStack Query
- Chart.js
- JWT + role-based auth
CRM Integration
- Kickserv API
- Customer sync, job booking
- Source tracking enum
- Priority tagging
Languages (hand-translated)
- English, Spanish, French
- Mandarin, Korean, Russian
- Hindi, Arabic
- Matched neural Polly voices
Ops & Quality
- Async outbound callback queue
- Retry + backoff logic
- Error-handling on every layer
- 98%+ dedup guarantee documented
Three productized layers of the same pipeline.
The phone agent that never sleeps.
Multilingual real-time voice agent on raw Twilio Media Streams + OpenAI Realtime API. Eight hand-translated languages with matched neural Polly voices. Pre-cached audio pipeline. Inbound calls handled in the customer's native language end-to-end. The front of the lead pipeline that drove +71.9% revenue lift at the anchor customer.
- Real-time streaming WebSocket protocol
- 8 languages hand-translated, neural-voice matched
- AI structured extraction on every call (priority detection, parts, duration)
- Auto-creates customer + opportunity in CRM
Web chat that deduplicates against your CRM.
GPT-4 backed chat that deploys onto the customer's existing site. Real-time qualification, appointment scheduling, IMAP email integration, OAuth. Writes deduplicated records into Kickserv with the same source-tracking and priority tagging as the voice agent. No site replacement required.
- Deploys onto existing sites (Hibu, WordPress, custom)
- OAuth + IMAP for email integration
- APScheduler for follow-up automation
- Same dedup engine as the voice channel
Operator command center for the whole pipeline.
React 19 + MUI v6 dashboard that unifies call logs, voice messages, customer records, settings, and cross-product workflow data. Real-time view of every lead as it lands from any of the five channels. Built so the operator's job becomes "show up to the booked appointment" instead of "field every call."
- Live lead feed from all channels in real time
- JWT + role-based access control
- Chart.js analytics: revenue, jobs, conversion
- TanStack Query + optimistic updates
The same shop. Other proof points.
AutoShop AI
Multi-tenant SaaS for mobile mechanics. FastAPI + PostgreSQL/pgvector + Redis/Celery. Self-serve onboarding, X-Tenant-ID isolation, hybrid IVR (ElevenLabs + Polly fallback), customer-facing PWA with offline support, Manifest V3 browser extension that auto-fills vehicle context across 7 parts suppliers.
War Library
Idea to live in 8 hours. Multilingual rollout days later (7 languages including RTL Arabic, Hebrew, Farsi). 1,799 page views, 13,492 verified events tracked across 169 countries, 18 monitored news sources. 129 automated tests. 8-parallel-Claude-Code-agent codebase audit producing 179 verified fixes. MIT licensed.
Operation: Orange Fury
Built in 6 days for Pieter Levels Vibe Jam 2026 while contracting full-time. #39 leaderboard finish with 5-min median playtime. 288+ plays. Single-file 9.7MB self-contained HTML with 31 PNGs base64-inlined, vanilla JS, PWA installable.
Wave-Field Cognition Framework
Mathematically formalized framework on EM field dynamics in neural computation. Pre-registered band-specific predictions, two-method convergence (Granger + KSG TE), public-dataset replication (OpenNeuro ds004306, DANDI 000623). Documented self-correction in changelog. Provisional patent US 63/839,719 (exploratory engineering, not a research credential).
AI Repair Pros
Native iOS and Android. React Native 0.72 + TypeScript, Clerk auth, Supabase, Redux Toolkit, react-native-community/voice for speech recognition, AdMob monetization, image crop picker.
Deutsche Auto Sport
My own German/European auto specialty operation in Fredericksburg. Built the public site solo. Live Lighthouse Desktop: Performance 90, Accessibility 94, Best Practices 100, SEO 100. I dogfood the Mid-Atlantic AI suite on my own SMB.
One engineer. End-to-end ownership.
Johnathan Scott Viruet
Production AI engineer shipping voice agents, multi-tenant SaaS, native mobile, civic tech, dashboards, and independent research. Four years compounding into a person who can take a customer outcome from problem statement to live deployment without handoffs — backend, frontend, infrastructure, multilingual content, integration, deployment, monitoring, customer relationship.
Fourteen years of customer-facing technical work in European automotive diagnostics before founding Mid-Atlantic AI in 2023. Customer-first framing is inherited, not learned. Methodology discipline (pre-registration, kill conditions, two-method convergence, documented self-correction) carries from the research work into the product work.
Two paths. Both real.
Hire me as a Forward Deployed Engineer.
Senior / staff Forward Deployed Engineer, Solutions Engineer, or Applied AI roles at voice AI infrastructure companies, AI-first startups, and frontier labs. Production experience with raw Twilio Media Streams, OpenAI Realtime API, multilingual NLP across 8 languages, multi-tenant SaaS, and customer-facing technical work. Remote-first preferred. Open to relocation for the right role.
Schedule a consult for your business.
Tell me about your business in plain language. Real person, same-day reply. If we can help, I'll say how. If we can't, I'll tell you who can. The case study above is the kind of result the engagement produces. Verifiable receipts available on request.