Skip to content
Mirun Kaushik
← All work

04Hackathon20253rd place, DevSpark 2025

Aura Mall

An AI shopping assistant that searches a mall by photo

My part
Hackathon team
Status
Built, with parts designed or planned
Stack
CLIP, RAG, MongoDB Atlas, Blender
Code
GitHub ↗on a teammate’s account

Fig. — How data moves through it

  1. 01Shopper’s photo
  2. 02CLIP embedding(model call)
  3. 03Similarity matchinventory image embeddings
  4. 04Similar items + outfit
  5. 05Route to storeBFS across floorsTeam
Model call Rules, data, queues Built by teammates

01Problem

In a large mall it is hard to find a product, know which store has it, and get there.

02Approach

Embed a shopper’s photo with CLIP and compare it against embeddings of inventory images. Questions about stores and products go to a RAG chatbot over the mall’s data.

03What I built

  1. 01Visual product search, like Google Lens but built from scratch: photograph an item, the photo is embedded with CLIP and matched against inventory image embeddings to show similar products.
  2. 02Outfit recommendations that suggest matching items, such as trousers and shoes for a shirt.
  3. 03A RAG chatbot that answers questions about the mall, stores and products, using a sample inventory dataset I created in MongoDB Atlas.

04Built by my teammates

  • A 3D model of the mall in Blender showing every store.
  • A breadth-first-search pathfinder for store-to-store routes across floors.

05Not built yet

  • plannedConnecting the chatbot to real store inventory systems.
  • plannedAn SOS feature that broadcasts an alert to everyone using the app.