DaleelBites (دليل بايتس) — Voice-First UAE Food Broker

Speak a craving. DaleelBites checks what's actually available right now across Talabat, Deliveroo and Noon Food, tells you the best option out loud, and takes you straight to the order page.

The problem

Any chatbot can recommend a restaurant — that's commodity advice. The actual friction is everything after: opening three delivery apps, checking which one has the dish, comparing prices, reading reviews to see if it's any good, and finally tapping order. That's five minutes of app-switching for a decision that should take one conversation.

The solution

1. You say what you're craving. A voice agent (ElevenLabs) has a short back-and-forth — delivery or dine-in, any preferences — while it works in the background.
2. It searches live, in parallel, across delivery apps and review sites (context.dev), extracting real menu prices, ratings, and actual customer reviews as it goes. No pinned demo data — it searches fresh every time.
3. It ranks by what matters: first, which places actually carry the dish; then reviews (weighted by how many, so 200 reviews at 4.6 beats 5 reviews at 5.0); price only breaks a tie.
4. It gives you a verdict, spoken and on-screen — the pick, a runner-up, the dish photo, why it won, anything to watch out for — and hands you straight to the order page.

Every number in that verdict traces back to a real fetch. If a source comes back empty, the answer is "nothing found," never a guess dressed up as one.

See it in action

final_dalrrl.1.mp4

Try it live

Web app	https://hack-build-2.vercel.app
Chrome extension	see apps/extension/README.md — load-unpacked, no store listing yet

How it's built

Piece	What it does
apps/api	FastAPI backend. Runs the research pipeline, ranks results, builds the verdict, holds user accounts.
apps/web	Next.js app. The voice console — talk or type, watch the verdict land.
apps/extension	Chrome extension (MV3). Same voice agent in a side panel, plus a "Compare" button injected into the delivery apps themselves.
docs/	Deep-dive docs — architecture, data contracts, vendor API notes, deployment guide.

Stack: ElevenLabs (voice), context.dev (live web search/extraction), FastAPI + SQLite (backend, accounts), Next.js (web), vanilla JS + esbuild (extension).

Run it yourself

Backend

cd apps/api
python -m venv .venv
.venv/Scripts/pip install -r requirements.txt   # Scripts → bin on macOS/Linux

Set CONTEXT_DEV_API_KEY in apps/api/.env (without it, the API still runs and returns clearly-labelled sample data instead of live prices).

.venv/Scripts/uvicorn app.main:app --reload --port 8000

Web app

cd apps/web
npm install

Set in apps/web/.env.local: ELEVENLABS_API_KEY, NEXT_PUBLIC_ELEVENLABS_AGENT_ID, and NEXT_PUBLIC_API_URL (point it at your local backend, or leave it unset to use the live one).

npm run dev

Chrome extension

cd apps/extension
npm install
npm run build

Then chrome://extensions → enable Developer mode → Load unpacked → select apps/extension. Full walkthrough (including the voice side panel and mic permission) in apps/extension/README.md.

Learn more

• docs/ARCHITECTURE.md — how the pieces fit together
• docs/CONTRACTS.md — the data shapes and anti-fabrication rules the verdict has to obey
• docs/VENDOR-CONTRACTS.md — what was actually verified about each delivery app's pages
• docs/DEPLOYMENT.md — every environment variable, and what needs setting up by hand
• docs/FOOD-FLOW.md — the product flow this was designed around