AI news · November 6, 2026

I built Dial Hanna for Meta's $1M hackathon: an AI assistant that picks up when shops can't

My Meta Global AI Developer Hackathon submit (Indie track, Build for Your World). Sparked by a Budapest bakery that didn't answer — a working phone agent in ten days, and how to join the waitlist.

~7 min read

Two weeks ago I wrote that I got into Meta's $1M AI hackathon. On 4 Nov I hit submit. This one is less of a how-to and more of a "here's what I made and why".

Watch the demo

Dial Hanna demo — submitted for Meta's hackathon (Indie track, Build for Your World).

Why I keep joining hackathons

Before this one I joined Replit's buildathon, and somewhere along the way I fell for the format. You start from an empty page, you pick the topic, and you have a hard deadline that kills overthinking. No client brief, no roadmap. Just "what would I build if I could build anything in ten days?"

That Replit entry was CommentCatalyst. I later moved away from Replit and cut my stack to ~$20/month — same lesson as this week: ship the one flow that matters, then keep the bill sane.

For Meta's hackathon my answer was: something people could actually benefit from — not a demo that looks good on stage and then sits in a folder. The quiet flex of the week: with Meta's models, features and apps, a working phone agent is a ten-day build.

The bakery that didn't answer

I live in Budapest, so Hungary is where I started — and the problem statement was literally Build for Your World. The spark was ordinary, which is why I trust it.

I wanted to order a cake from a local bakery. I sent an email. No reply. I called. Nobody picked up. Two days later the email finally came back. I understood they were busy — the floor was full, the kitchen was flying — but in the meantime I had already started looking at other options.

That gap is the product. Bakeries, and a lot of other small businesses, lose money when they can't answer the phone or incoming queries promptly. The work walks next door. That's why Hanna is here.

Who Hanna is for

On Dial Hanna the starting templates cover the trades where hands stay busy and the phone still rings:

• Salon & beauty — scissors in hand, client in the chair
• Auto & tyre service — car on the lift, hands oily
• Trades & handymen — on a ladder, drill in hand
• Moving & cleaning — mid-move, can't pick up
• Guesthouses & apartments — cleaning, key handover, guest at the gate
• Trainers & studios — mid-session, phone away
• Photographers & events — shooting, not answering
• Pet grooming — wet hands, barking client on the table
• Hospitality — full floor, cake or table orders on the line
• Flowers & gifts — peak-season counter, phone ringing
• Language schools & tutors — in a lesson, phone in the bag
• Freelancers — in a meeting or in focus, phone on silent

Don't see your trade? The templates are starting points — Hanna adapts to any small business. If you need a more tailored process (your own systems, special notifications, custom booking rules), reach out through Hanna and we're glad to help connect her to your tools.

How Dial Hanna works

From the shop owner's side — the product flow, not the model IDs:

1. Keep your own number. No new line for callers. Turn on call forwarding when you're busy, in a meeting, training, or mid-job.
2. Paste your website URL. Muse Wonder reads it and drafts the knowledge base — business profile, opening hours, and the most likely customer FAQs — ready for you to approve. You extend the Q&A list; that list is Hanna's database. There's a verification call before go-live so forwarding really reaches her.
3. When you're unavailable, Hanna answers. She introduces herself as an AI in the first sentence (EU Article 50). She takes multi-slot orders and bookings over the phone — product, size, pickup time, name, phone — reads them back for confirmation, and closes the call politely. After every call the owner gets Intent → Outcome → Next action, not a transcript dump.

When Hanna doesn't know: knowledge gaps become FAQs

If Hanna doesn't know something, she doesn't guess. She offers a callback, logs a knowledge gap, and the owner answers once in the dashboard. That answer becomes an approved FAQ, so the next caller gets it instantly. The knowledge base grows from real calls — product learning, not a second copy of the stack table.

The Meta stack (and one voice layer)

I built Dial Hanna on Meta Muse Code / Meta Model API (api.meta.ai/v1). Three Meta models do the agent work; ElevenLabs Flash is just the spoken voice — the program has no Meta TTS. Model IDs live here:

Honest limits I'm not papering over:

• Speech-to-text is English for now. Muse Voice Transcribe has no Hungarian STT in this build yet. Hanna's spoken voice can still be Hungarian or English — hearing the caller in Hungarian is the open gap.
• Muse Spark has to actually call tools for booking and order slots — that's why it's the call brain, not just a chat model.
• One Meta API for Wonder, Spark and Transcribe. ElevenLabs only says the words out loud.

The flip side of a ten-day Muse build

I'm not going to pretend the week was smooth. A few things hurt:

1. Muse Code in the terminal can be very slow on complex specs and bigger changes. When the prompt got thick, I waited a lot.
2. The demo video — render plus transcript — took hours. Everything else was ready during the day on 4 Nov; I hit submit around 10–11 PM Budapest time, only a couple of hours before the final deadline (4 Nov, 9:00 PM ET / 5 Nov, 03:00 Budapest).
3. Long-context loss on Muse Spark 1.3. In a long session it lost prior conversation, specs and markdown files and went a different direction. I had to steer it back. On the hackathon feedback form I also logged the biggest obstacle as the model not following instructions — useful honesty if you're planning a similar sprint.

None of that kills the thesis — Meta's stack still got me a working phone agent in ten days — but it's the honest flip side.

How I'd pick models after this week

Verified list prices as of 6 Nov 2026:

Wonder and Opus sit in the same dollar band at $4/$20. Argon is on intro pricing until the end of November, then joins that $4/$20 class. Full Muse Wonder breakdown: Muse Wonder: Meta pricing, API and benchmarks. My take after this week (builder experience, not a ranking law):

• Claude Opus 5.5 currently leads for me on coding-agent autonomy and terminal use. In the Dial Hanna week I did not see the same context-loss / slow-reply pattern with Opus 5.5 that I hit with Muse Spark 1.3 in a long Muse Code session.
• Gemini 4 Argon currently leads for me when the job is multimodal or document retrieval.
• Muse Wonder is mainly useful when you're inside Meta workflows, frameworks and Meta products (Facebook, Instagram, WhatsApp) — which is exactly why it fit a Meta hackathon phone agent.

For Dial Hanna I stayed on Muse because the brief was Meta. For day-to-day coding agents outside that lane, Opus 5.5 is where I'd start right now. (If you're picking a $20 coding plan, I compared them in Which AI coding tool should you pay $20 for?.)

Where it stands

• Waitlist. Join at dialhanna.com/waitlist.
• Track. Indie Developer · Build for Your World · team Vibe Coder's Life.
• Hungary first, world second. Site default is HU, with an /en toggle; Hanna is built for shops here, then further out.
• Next personal milestone: getting the first real customer on board. Hackathon judging runs about 4–8 Nov — winners around 9 Nov.

What I'd tell you if you're thinking about your first hackathon

• Pick a problem you watched happen. Mine was a Budapest bakery that couldn't answer while I waited on a cake.
• Keep the plan tight. You still have to deliver a finished product by the deadline — scope is the whole game.
• Budget time for the canvas, videos and demos. Render + transcript took me hours; leave more buffer than you think you need.
• Watch how the deadline is structured. Some buildathons split into plan / build / publish with separate stage deadlines. Meta's Global AI Developer Hackathon this time was a one-off submission deadline — one shot, not a staged runway.
• Lean on the models you're given. Muse Wonder for the website, Muse Spark for the call, Muse Voice Transcribe for hearing, Muse Code to wire it — that split made the sprint possible.

Related on Vibe Coder's Life

• I got into Meta's Global AI Developer Hackathon 2026: dates, credits, eligibility — part one of this story
• Muse Wonder: Meta pricing, API and benchmarks vs Opus 5.5 and GPT-6 Astra — the model behind Hanna's knowledge base
• Muse Code vs Claude Code and Cursor — Meta's pay-as-you-go coding agent — the Muse Code map I used before this sprint
• How I moved away from Replit and cut my stack to ~$20/month — the stack path after my earlier buildathon
• Which AI coding tool should you pay $20 for? — where Opus, Gemini and friends land for everyday coding
• How I built Grok Mail — another build-in-public story

FAQ

What is Dial Hanna?

Dial Hanna is an AI phone assistant for small businesses, built for the Meta Global AI Developer Hackathon 2026 (Indie Developer Track, Build for Your World). You keep your own number; when you're busy, call forwarding reaches Hanna. She answers from owner-approved knowledge built from the shop's website, takes multi-slot orders and bookings, leaves Intent → Outcome → Next action on the dashboard, and grows the FAQ list from real knowledge gaps. As of November 2026 join via dialhanna.com/waitlist.

Which hackathon track did you submit to?

Indie Developer Track, problem statement Build for Your World. Team name Vibe Coder's Life. Submitted 4 November 2026.

Why did you build Dial Hanna?

Because I tried to order a cake from a Budapest bakery: the email sat unanswered, nobody picked up the phone, and by the time they replied two days later I was already looking elsewhere. Bakeries and many other small businesses lose work when they can't answer promptly. Hanna is the assistant who picks up so that work doesn't walk away.

Who is Dial Hanna for?

Starting templates on Dial Hanna cover salons, auto and tyre shops, trades, moving and cleaning, guesthouses, trainers, photographers, pet grooming, hospitality, flowers and gifts, language schools, and freelancers — but Hanna can adapt to any small business. If you need a bespoke process, reach out through Hanna.

Can an AI answer my business phone when I'm busy?

Yes. With Dial Hanna you keep your own number and forward busy or unreachable calls. Hanna introduces herself as an AI, answers from the approved knowledge list, takes notes and orders, and can book appointments. She shouldn't guess outside that knowledge — she offers a callback instead.

Can Hanna take orders and book appointments?

Yes. In this build she handles multi-slot orders and bookings over the phone (for example product, size, pickup time, name and phone), reads them back for confirmation, and can book into Google Calendar. Anything outside the approved knowledge should go back to you.

Which models does Dial Hanna use?

Muse Wonder builds the website knowledge base. Muse Spark runs the call (answers, orders, Intent/Outcome/Next). Muse Voice Transcribe turns speech into text — English only for now. Hanna's spoken voice is ElevenLabs Flash. The agent was wired with Meta Muse Code on Meta Model API.

Do AI phone agents have to say they're AI in the EU?

Yes. Since 2 August 2026, Article 50(1) of the EU AI Act requires AI systems that interact directly with people to tell them they're dealing with an AI, unless it's obvious. The European Commission says this must happen from the start of the first interaction. Hanna does it in the first sentence of every call.

Can I try Dial Hanna now?

Yes — open dialhanna.com. The homepage has a sample conversation under See how it works. For the waitlist, join at dialhanna.com/waitlist.

When are the Meta hackathon winners announced?

Judging runs from about 4 to 8 November 2026, and winners are announced around 9 November 2026. The prize pool is $1,000,000 across two tracks: indie developers and startups.