KadakSpot

The AI app for foodies :)

Comment

Inspiration

Finding food shouldn’t feel like work. Most apps flood you with endless lists and generic ratings. We wanted something that feels human: tell it what you’re craving, see what’s actually good near you, and go. That’s how KadakSpot was born—AI that speaks your language, grounded in real local signals.

What it does

KadakSpot is a hyperlocal food discovery app with three pillars:

Smart Map:

Live map with location permission, quick Restaurant/Cafe modes, and a Qloo AI toggle. Turn it on to see a heat map of rising/popular spots around you.

LLM Chat:

Ask “healthy options near me?” and get grounded recommendations that blend Qloo intelligence with Google Maps/Places data (names, distance, hours).

Power Settings:

Dark mode, map online/offline behavior, and chat options (save logs, pick on-device vs. off-device models).

How we built it

Frontend:

React Native + Expo (TypeScript). Clean UI, snappy transitions, sensible loading states.

Maps: Google Maps/Places APIs for geocoding, nearby search, place details; custom overlays for heat map + pins.

Taste Intelligence:

Qloo API for culture/taste graph–based recommendations and trending signals.

AI Layer:

Chat flow that fuses Qloo + Maps results into coherent, actionable answers (debounced input, streaming updates).

UX/Settings:

Theme switch, connectivity controls, and configurable chat storage/model selection to respect privacy and performance.

Challenges we ran into

Grounding the LLM:

Preventing “pretty answers” that don’t map to real places. We fixed this by always attaching IDs/coords from Google and cross-checking with Qloo.

Rate limits & latency:

Batched requests, caching, and optimistic UI to keep the app responsive.

Cross-platform maps quirks:

Permissions, web/mobile differences, and rendering performance for pins + heat map.

Search intents:

Normalizing free-form queries (“tasty burger” vs. “cheap biryani”) into consistent filters.

Accomplishments that we’re proud of

  • A smooth map → chat → settings flow that actually reduces decision fatigue.
  • Qloo heat map overlay that surfaces rising/popular spots at a glance.
  • Grounded chat answers that include real place cards you can tap and route to.
  • Flexible settings: on-device model option, offline-friendly map behavior, and neat dark mode.

What we learned

People want fewer, better choices—not 200 pins. Ranking and context matter more than raw volume.

LLMs shine when grounded in real data (IDs, coordinates, hours) and bounded by sensible filters.

Tiny UX touches (progress indicators, debounced search, card previews) dramatically improve trust and speed.

What’s next for KadakSpot

Personalization:

Lightweight taste profiles, “teach KadakSpot your vibe,” and feedback-driven reranking.

On-device intelligence:

Smaller local models for privacy + low latency.

Richer signals:

Live busyness, price bands, dietary tags, and time-of-day aware suggestions.

Social & lists:

Shareable shortlists, “friend picks,” and collaborative planning.

Merchant tools:

Claim a spot, add specials, and run taste-aligned promos.

Built With

Share this project:

Updates