← All Projects
AI & Machine Learning2026

Horizontal AI Receptionist for SMBs

Venture: Callen AI

Callen AI is a horizontal AI receptionist platform for SMBs in Southern Europe, launching through the dental vertical in Greece. Inbound calls are answered, transcribed, and converted to bookings against the practice's calendar — self-serve, live in 30 minutes, no integrator required. The wedge against enterprise-focused incumbents like Voice Logica is speed, simplicity, and SMB-tier pricing.

Key ResultProduction-ready voice pipeline: inbound calls answered, transcribed, qualified, and booked against the practice's Google Calendar — fully self-serve, live in ~30 minutes with no integrator
Engagement
In-house venture by WeAreFabbrik
Timeline
~14 weeks to ~90% production-ready
Team
  • Konstantinos Tsolakidis — Founder & Product
  • Marios Tasos — Senior Full-Stack Engineer
  • Nikiforos Kolaros — UI/UX & Branding
Scope of work
Product strategy & roadmap · Full-stack build · Voice AI pipeline (Twilio + ElevenLabs + GPT-4) · Self-serve onboarding · Billing & metering · Scheduler & call analytics · Security / middleware hardening
Services
AI & Machine Learning · Custom Platforms · Business Automation
Stack
TypeScript · Next.js · Node.js · Supabase · Twilio · ElevenLabs · OpenAI GPT-4 · Google Workspace API · Stripe · Vercel

Challenges

  • Building a voice AI product an SMB owner can adopt without procurement, integrators, or a multi-week deployment — the empty middle between consumer demos and enterprise platforms
  • Keeping the platform horizontal in architecture while going to market like a vertical specialist — dental as a beachhead, not a destiny, with clean sequencing into clinics, law offices, restaurants, and real estate
  • Differentiating against an enterprise-focused incumbent (Voice Logica, partnered with telecom Yuboto) on speed, simplicity, and SMB pricing rather than feature parity

Solutions

  • TypeScript end-to-end on opinionated, boring infrastructure — Next.js on Vercel, Node + Supabase Postgres, Twilio + ElevenLabs + OpenAI GPT-4 voice path, Google Workspace for calendar and email, Stripe for billing. Restraint as the product.
  • Five-step self-serve onboarding with in-app Twilio number purchase, opinionated dental assistant templates, JWT + httpOnly auth, RBAC, rate limiting, webhook signature verification, and a single dashboard surfacing calls, emails, calendar, and follow-ups
  • Tiered per-number metered subscriptions (€49 Starter / €99 Professional / €199 Business) plus a one-time €997 Founder's Circle lifetime deal capped at the first 10 customers — a forcing function for trust pre-launch and a permanent reference cohort

Results

Shipped
  • Production-ready voice pipeline: inbound calls answered, transcribed, qualified, and booked against the practice's Google Calendar — fully self-serve, live in ~30 minutes with no integrator
  • Full platform shipped: five-step onboarding with in-app Twilio number purchase, billing & metered subscriptions, scheduler, admin panel, call analytics, AI training surface, and a hardened middleware/security stack (JWT + httpOnly auth, RBAC, rate limiting, webhook verification)
Validated
  • Horizontal architecture and vertical go-to-market proved out: one opinionated stack serving a dental beachhead, with clean sequencing into clinics, law offices, restaurants, and real estate
  • Positioning learning: emotional-pain copy ("the patient who didn't call back") meaningfully outperformed cost-savings / ROI copy on conversion intent
  • Pricing thesis validated: per-number metered tiers (€49 / €99 / €199) plus a €997 lifetime Founder's Circle as a pre-launch trust forcing-function
Targets — next 12 monthsGoals, not achieved results.
  • First 10 Founder's Circle customers locked in as design partners and a permanent reference cohort
  • 50 paying SMBs and a credible second vertical within 12 months of launch
Every missed call in a Greek dental practice is somewhere between thirty and two hundred euros walking out the door — and the dentist almost never sees the bleed because it happens while they're elbow-deep in someone else's mouth.
Konstantinos Tsolakidis · Founder & Product, WeAreFabbrik

The Problem

Small businesses in Greece — and across Southern Europe — are losing customers every single day because nobody can pick up the phone, and the existing solutions are either too expensive, too complex, or too generic to fix it.

A solo dental practice in Athens takes 40 to 70 inbound calls a week and misses roughly 30% of them — during procedures, lunch, end of day, weekends. Each new patient is worth €300 to €1,200 in lifetime value. The math is brutal: a single missed booking covers a whole month of the Professional plan. But the dentist doesn't feel the math — they feel the patient who said "I'll just call the next clinic on the list." That's an emotional wound, not a spreadsheet line. We chose to build for that wound.

The problem has been "solvable in theory" for two or three years — LLMs are good enough, voice synthesis is good enough, telephony APIs are mature. What hasn't existed is a product an SMB owner can adopt without a procurement process, an integrator, or a six-week deployment. The market is full of consumer-grade demos and enterprise-grade platforms. The middle is empty, and that middle is where Greek SMBs live.

The Approach

The hardest decision wasn't a technical one. It was deciding to keep the platform horizontal in the architecture and the vocabulary, while going to market like a vertical specialist. Most teams pick one or the other. Both is harder, but it's the only framing that survives a five-year horizon.

The bet is that the long-term winner in AI receptionists won't be the company with the deepest dental features or the most enterprise integrations — it'll be the company that owns SMB onboarding. We're building the Stripe of voice agents for small businesses: opinionated defaults, sane pricing tiers, self-serve setup, and a deliberately narrow surface area.

Greek dental was chosen as the beachhead because the call patterns are clean, the value-per-call is high, the pain is acute, and one satisfied dentist generates word-of-mouth across a whole neighborhood of clinics. The platform is horizontal, with clear sequencing into clinics, law offices, restaurants, real estate, and tourism.

Voice Logica leads with feature breadth and enterprise credibility. Callen leads with emotional pain. Their hero copy talks about omnichannel automation. Ours talks about the patient who didn't call back. Their sales motion is a quote and a discovery call. Ours is a 14-day free trial and a Stripe checkout. The wedge against an incumbent is never feature parity — it's a different shape of product entirely.

Built in-house — and run in production

Callen AI isn't only a client engagement — it's a product we conceived, built, priced, and took to market ourselves. That's deliberate proof: when you hire WeAreFabbrik you get a team that has shipped a live SaaS end-to-end — voice infrastructure, billing, onboarding, go-to-market — not just a team that writes code to spec.

What We Built

The boring answer is the right one. The interesting part is what we chose not to build.

TypeScript end-to-end. React/Next.js on Vercel, Node.js backend, Supabase Postgres, JWT plus httpOnly cookies for auth, and a hardened middleware layer covering rate limiting, RBAC, input validation, and webhook signature verification. The voice path uses Twilio for telephony, ElevenLabs for synthesis, and OpenAI GPT-4 for intent. Email and calendar run through Google Workspace APIs. Stripe handles both metered subscriptions and the one-time Founder's Circle payment. A cron-driven scheduler runs daily briefings, email summaries, and overdue follow-up flagging.

A dentist signs up, runs the onboarding wizard, picks a Greek phone number from inside the app, configures an AI assistant from a dental template, and forwards their existing line. From that point forward, every inbound call is answered by Callen — the agent qualifies the caller, books or reschedules an appointment against the dentist's Google Calendar, transcribes the conversation, and creates a follow-up task if anything was left open. The dentist sees all of it on one dashboard the next morning, alongside an AI-summarized digest of their email inbox and the day's calendar.

What we deliberately did not build is the product. No custom telephony layer — Twilio wins that fight. No proprietary voice model — ElevenLabs is better than anything we'd ship internally. No CRM — we integrate with the calendar and inbox the dentist already uses. No per-customer custom integrations — every dentist gets the same opinionated stack, and that's a feature, not a limitation. The closest thing to "custom" the customer ever encounters is an assistant prompt template they can lightly edit. That restraint is the product.

Pricing & Go-to-Market

Pricing is per phone number with a metered-minute component on top — usage scales with value rather than seat count. A small practice running one number pays Starter pricing forever if that's what fits.

Starter€49/mo

Single number, solo practice, low call volume.

Business€199/mo

Multi-number, established multi-location practice.

EnterpriseCustom

Multi-location, high volume, custom-quoted.

Founder's Circle€997 lifetime

One-time payment, capped at 500 minutes of AI handling per month, hard-limited to the first 10 customers, non-transferable. A forcing function for trust at pre-launch and a permanent reference cohort.

Lessons

  • Lead with emotional pain, not rational savings

    The first round of website copy was built around cost-savings math: hours saved, headcount avoided, ROI. Technically correct, emotionally inert. The second round leaned into the missed call, the lost patient, the dentist who finds out at the end of the day that three people called and nobody picked up. Conversion intent on the new copy is meaningfully better. Rational arguments justify a purchase; emotional pain triggers it.

  • Restraint is a feature

    Every time a configuration knob, an edge-case branch, or a "we can also do X" option got added, onboarding completion got worse. The version that ships in week one is deliberately narrower than the version the team knows how to build. Templates beat custom prompts. One dashboard beats five. Opinionated defaults beat exposed settings.

  • Buy the boring infrastructure

    The Twilio bill, the ElevenLabs bill, the OpenAI bill — those are operating costs paid happily because the time saved by not building them is the entire reason the product ships in months instead of years. Custom code only lives where it touches the customer directly: onboarding, assistant configuration, call-routing, the dashboard.

  • Naming is positioning

    The original company name — "AI Secretary" — described what the product did but not what the customer felt. Renaming to Callen AI was a conscious move from functional naming to brand naming, and it unlocked positioning the old name couldn't reach.

What's Next

The four-to-six-week path to launch covers Google OAuth token refresh, error boundaries, assistant editing, and a first pass of test coverage. Public pricing goes live with the launch; the Founder's Circle window opens the day after. Longer-term, the roadmap sequences a second vertical by month nine, multi-language assistants by month twelve, and an embedded-partner channel into year two.

Technologies Used

TypeScriptNext.jsNode.jsSupabaseTwilioElevenLabsOpenAI GPT-4Google Workspace APIStripeVercel

Have a Similar Project?

Let's discuss how we can build something great together.

Built with our AI & Machine Learning and Custom Platforms practice — that's what we'd bring to your product.