Logo Lanfrica

abrishdew/LLM-Finetuning_Generate-Amharic-based-creative-text-Ad-contents

Domain:

natural language processing

Record type:

project
Creator:
abr
Host:
Builds an Amharic text generation system for targeted Telegram ads. It fine-tunes a Large Language Model (LLM) on Amharic text for high-quality generation. A Retrieval-Augmented Generation pipeline considers campaign data and channel history to improve ad creativity and relevance, leading to more effective advertising in the Amharic-speaking # LLM Finetuning: Generate Amharic text based creatives /Ad contents/ *** ### Project Overview An startup in Africa focused on AI and Blockchain business solutions needs to adjust its advertising strategy to better fit this ever-changing ecosystem - in our case Telegram messaging app. For this project, We have the intention to improve the effectiveness of promotional efforts by integrating powerful AI capabilities for Amharic text manipulation, in particular creating an Amharic RAG pipeline that will help generate Amharic based creative text Ad contents given campaign information such as brief (brand and product information) and the content history of a telegram channel. This project makes sure that their advertisements are both catchy and relevant to the Telegram community. To achieve this, it is required to have quality **amharic text embedding** and **text generation**. ### Project Guide After choosing a suitable open source LLM model with appropriate training and one that should already have a capability to embed amharic texts, a dataset will be utilized to finetune it further to deliver the business needs. That is made possible by using an efficient RAG system to help generate ideal Ad contents in Amharic. The data contains exported Telegram messages in JSON format representing 25 telegram public channels. The structure of the JSON object provides information about the sender, timestamps, message types, and any associated multimedia content. In the end, A RAG pipeline will be set for the final Ad generation which will be hosted on React. ### Approaches and Techniques The data seems to be a dump and needs to be cleaned and preprocessed so as to get the desired and ideal dataset to finetune the open source model. THe finetuning will take place in two steps. It will be used to train the base model so it is possible to enable it to know more Amharic Context, and then we will have a filtered and labeled dataset (filtered with Ad context only) to help it underst …