10 Academy profile site

Developed a Retrieval-Augmented Generation (RAG) system for precise contract analysis, improving accuracy and efficiency in answering legal queries. Implemented text chunking, semantic search, and tested different embedding models to enhance retrieval and response generation. Evaluated the system using RAGAS metrics, ensuring accuracy and precision in the system.

Fine-tuned a Llama 2 language model for Amharic news headline generation, utilizing web scraping, Kafka, Faust, and MongoDB for scalable data preprocessing. Addressed data scarcity challenges and optimized performance with techniques like parameter-efficient fine-tuning, contributing to enhanced Amharic NLP capabilities.

A Redash chatbot streamlines data exploration for non-technical members, transforming complex SQL queries into natural language conversations. This empowers non-technical stakeholders to directly access and analyze key business metrics within Redash and make informed decisions without needing SQL expertise. 

This project developed an automated prompt tuning system for enterprise RAG systems. Leveraging LangChain and OpenAI's language models, it streamlined the generation, evaluation, and ranking of prompts to enhance the accuracy and relevance of RAG-based applications. It generates diverse test cases, creates candidate prompts, evaluates their performance, and provides a user-friendly interface for selecting the most effective prompts.