Yesterday, deciding what to eat could mean missing your chance to order.
We're solving that with Dine.
A Telegram bot that plans your meals, automates payments and places orders for you.
Follow as I build with @munachi_nw@thekorahq@korabytescu#Korabytes#Kora#BuildInPublic
Wrong item name? Stop. Cart not empty? Stop. Our ordering agent checks for exact matches and refuses to proceed when conditions aren't right. When money is involved, failing safely beats guessing.
@thekorahq@korabytescu#Korabytes#Kora#BuildInPublicChallenge#Dine
Wrong item name? Stop.
Cart not empty? Stop.
Our ordering agent checks for exact matches and refuses to proceed when conditions aren't right.
When money is involved, failing safely beats guessing.
@thekorahq@korabytescu#Korabytes#Kora#BuildInPublicChallenge#Dine
What happens if the AI goes down?
Dine falls back to keyword search, returning real meals, prices, and allergen warnings.
The bot should remain useful even when the model isn't available.
@thekorahq@korabytescu#Korabytes#Kora#BuildInPublicChallenge
AI can recommend a meal, but should it be able to invent its price?
As we build Dine, we're ensuring recommendations are grounded in verified catalogue data, not AI generated assumptions.
@thekorahq@korabytescu#Korabytes#Kora#BuildInPublicChallenge
Dine combines AI with deterministic logic.
AI interprets dietary requirements; backend rules enforce them during meal selection.
Dietary restrictions shouldn't depend on an LLM's memory.
@thekorahq@korabytescu#Korabytes#Kora#BuildInPublicChallenge
What happens if the AI goes down? Dine falls back to keyword search, returning real meals, prices, and allergen warnings. The bot should remain useful even when the model isn't available.
@thekorahq@korabytescu#Korabytes#Kora#BuildInPublicChallenge
A payment notification isn't proof of payment. When a student funds their Dine wallet, we verify Kora's webhook signature and confirm the transaction directly with Kora before crediting the wallet.
@thekorahq@korabytescu#Korabytes#Kora#BuildInPublicChallenge#Dine
If you ask GPT or Gemini: "Give me dinner for ₦2,000 with 45g of protein and no peanuts," it will happily generate a meal that costs ₦3,400, contains oil, and cites a restaurant that closed last month.
Dine solves it.
@thekorahq@korabytescu#Korabytes#Kora#BuildInPublic
2. AI Intent Extraction
Students don't speak in database queries. We use Gemini to translate natural-language requests into structured food preferences, with a regex fallback when the API is unavailable.
Production-grade Telegram bots need more than a bot.ts file.
For Dine, we built a modular TypeScript monorepo using pnpm workspaces, AI intelligence, ordering workers, database, and shared contracts.
@thekorahq@korabytescu#Korabytes#Kora#BuildInPublic
When we set out to build Dine, we didn't want to build another delivery app with 500 options.
Students want quick affordable choices.
We asked: What if an AI could act as your personal food assistant?
More 👇
@thekorahq@korabytescu#Korabytes#Kora#BuildinPublicChallenge
You give it your budget (say, ₦2,000).
You tell it your goals ("bulking / high protein", "post-workout", or "recovering from purging").
You drop a raw voice or text note: "Allergic to groundnuts, no amala, nothing fried."
It formulates an optimal meal.
We're combining AI with deterministic logic in Dine.
AI interprets requirements. Backend rules enforce it during meal selection.
A dietary restriction shouldn't depend on whether an LLM remembers to follow it.
@thekorahq@korabytescu#Korabytes#Kora#BuildInPublicChallenge
What happens when a user tells your AI to ignore its instructions?
We're testing how Dine handles prompt injection, misleading prices.
Building AI means testing its limits, not just its capabilities.
@thekorahq@korabytescu#Korabytes#Kora#BuildInPublicChallenge
To get good data for dine's intelligence, we sent out a form and in the first hour, over 100 students have shared their food experiences, reviews, and feedback.
Haven't filled the form... link is in the comments
@thekorahq@korabytescu#Korabytes#Kora#BuildInPublicChallenge
Good AI starts with good data. We're collecting and structuring food reviews, ratings, and student feedback to help train Dine's AI to make smarter, personalised meal recommendations.
@thekorahq@korabytescu#Korabytes#Kora#BuildInPublicChallenge