---
title: "Discussion on \"How I Built a RAG Assistant for a Pakistani University with Next.js, Convex and Groq\" | Hashnode"
description: "Discussion on \"How I Built a RAG Assistant for a Pakistani University with Next.js, Convex and Groq\". The architecture behind UETGPT: hybrid search, contextual chunks and answers that cite official sources."
url: "https://hashnode.com/posts/how-i-built-a-rag-assistant-for-a-pakistani-university-with-next-js-convex-and-groq/6ac738b601c233bdd8b51e30"
---

[Feed](https://hashnode.com/feed) /

Discussion

# [How I Built a RAG Assistant for a Pakistani University with Next.js, Convex and Groq](https://ibrahimsalman.hashnode.dev/how-i-built-a-rag-assistant-for-a-pakistani-university-with-next-js-convex-and-groq?utm_source=hashnode&utm_medium=feed)

The architecture behind UETGPT: hybrid search, contextual chunks and answers that cite official sources.

Oct 8 · 5 min read · ibrahimsalman.hashnode.dev  
Read the article

-   [#rag](https://hashnode.com/tag/rag)
-   [#nextjs](https://hashnode.com/tag/nextjs)
-   [#ai](https://hashnode.com/tag/ai)
-   [#convex](https://hashnode.com/tag/convex)
-   [#build-in-public](https://hashnode.com/tag/build-in-public)

## Written by

[Ibrahim Salman](https://hashnode.com/@devhms)

@devhms · Islamabad, Pakistan

  
[View profile](https://hashnode.com/@devhms)

Senior software engineer building AI tools. Flagship: UETGPT (uetgpt.site). Portfolio: ibrahimsalman.vercel.app

I'm a senior software engineer in Islamabad, building AI tools that solve real problems. My first flagship project is UETGPT (https://uetgpt.site), a free, unofficial AI guide to UET Taxila admissions that answers from official sources and cites them. I have more projects live and more in progress. See them at https://ibrahimsalman.vercel.app

## Responses 1

Markdown supported · ⌘↵ to post

[Ahmet Özel](https://hashnode.com/@ahmetozel)

[Oct 11](https://hashnode.com/posts/how-i-built-a-rag-assistant-for-a-pakistani-university-with-next-js-convex-and-groq/6ac738b601c233bdd8b51e30/comment/6acb5b11b19fbe5ad15d6a72)

The parent-child retrieval design is helpful for prospectus tables, but parent expansion should preserve the child's exact supporting row as well as the surrounding section. Otherwise a relevant fee row can be buried among several categories again when the larger parent reaches generation.

I would keep a row identifier and page location with each child hit, then show which fields the answer used. A test with two adjacent fee categories and one amount split across a page boundary would check the extraction repair and the expansion together. That makes the citation inspectable at the same granularity as the number being answered.
