← Back to work
Amharic RAG Ad Builder
RAGHugging FaceAmharic NLPFastAPI

Amharic RAG Ad Builder

2025Training project

A training project that uses RAG and open-source LLMs to generate Amharic text ads for Telegram channels.

The problem

Most generative-AI tooling assumes high-resource languages, which leaves Amharic underserved.

How it works

  1. 01, Input
    Knowledge base

    Reference content the ads are grounded in.

  2. 02, Tool or code
    Retrieval

    Finds the relevant context for each ad.

  3. 03, Model
    Open-source LLM

    Generates the ad text in Amharic.

  4. 04, Output
    Telegram ad copy

    Context-aware Amharic ads for Telegram channels.

InputTool or codeModelOutputServed through FastAPI.

What I did

  • Built an Amharic RAG pipeline using open-source LLMs.
  • Generated contextually relevant Amharic text ads for Telegram channels.
  • Focused on making generative AI work well for a low-resource language.

Results

  • Generated Amharic ad copy using retrieved knowledge-base context.
  • Explored RAG behavior in a low-resource language setting.
View code ↗
← All projects