Portfolio demo. ShopEase is a fictional store, so nothing here is for sale. The support assistant on the right is the real project. How it works

SUPPORT ONLINE · REPLIES IN SECONDS

Hi, how can we help?

Ask about an order, delivery, payment or return. The assistant looks up ShopEase's own handbook before every answer, so it won't make up a policy.

Prefer a person? WhatsApp 0300-1234567 or email support@shopease.pk, every day 9 AM to 10 PM.

ShopEase support

Answers from the store handbook

LIVE TRACE
Can I return earbuds if I don't like them?
Earbuds can't be returned once the seal is broken, unless they arrived damaged or wrong. In that case, report it within 48 hours with photos.

1.1s · 812 in / 38 out

Browse help topics

Pick a question and the assistant answers it straight away.

6 TOPICS · 9-PAGE HANDBOOK

RETRIEVAL-AUGMENTED GENERATION

How the assistant answers

The model never memorised the store's rules. It looks them up in the handbook before every answer, and says so when it can't find one.

  1. 01 / INGEST

    A PDF handbook

    Store policies live in a 9-page PDF, not in the model.

  2. 02 / EMBED

    Chunks and vectors

    LangChain splits the PDF; a local model turns each chunk into a vector stored in PostgreSQL with pgvector.

  3. 03 / RETRIEVE

    Semantic search

    Your question becomes a vector too, and the closest chunks by meaning are pulled back.

  4. 04 / GENERATE

    Grounded answer

    Claude answers only from those chunks, so it never invents a policy.

  • Next.js
  • Flask
  • LangChain
  • PostgreSQL + pgvector
  • Sentence Transformers
  • Claude
  • Docker