Machine Learning Engineer
Gebeya Inc.
Developed and deployed a suite of AI-powered services for a talent marketplace, focusing on enhancing talent matching, data analysis, and user interaction through a microservices architecture.
Key Achievements:
- 1Enhanced AI talent search for a SaaS platform using Generative AI and RAG, incorporating an SVM CV classifier and semantic resume ranking, projected to save clients tens of recruitment hours.
- 2Designed an agentic AI chatbot (LangGraph, Autogen) with GPT-4 function calling and multi-channel integration, streamlining engineering team access to AI tools and boosting internal productivity.
- 3Implemented Langfuse for comprehensive AI interaction tracing and Redis caching, optimizing system performance, and contributing to a notable reduction in API operational costs.
- 4Advanced algorithm development for candidate matching using Python and embedding models, achieving a 90% accuracy rate and elevating average client satisfaction scores by 20 points.
- 5Led R&D for an autonomous video subtitling system (English to Amharic), delivering over 90% accuracy in automated subtitling and documenting key AI dubbing challenges.
Products & Projects
AI-Powered Talent Matching & Vetting
AI-powered talent matching and vetting system for a SaaS platform, using Generative AI and RAG, incorporating an SVM CV classifier and semantic resume ranking, projected to save clients tens of recruitment hours.
Throughout Gebeya Saas product I was able to work on - Document Classification: Engineered a high-accuracy CV/Resume classifier by training an SVM model on a dataset of over 100,000 scraped and internal documents, outperforming initial keyword-based and early GPT-3.5 approaches. Semantic Resume Ranking: Developed a sophisticated ranking system using the 'all-mpnet-base-v2' model and experimented with multiple open-source embedding models (Jina, BGE, GTE). Implemented a Semantic-Aware Text (SAT) algorithm and tested various similarity metrics (Cosine, Euclidean) to rank resumes based on contextual relevance, significantly improving match quality. Contextual Summarization: Fine-tuned BERT and Longformer models to generate concise summaries of resumes and job descriptions, overcoming short context window limitations.
Agentic AI Chatbot through Multi-Channel Integration
An agentic AI chatbot through multi-channel integration for Gebeya SaaS platform that enables users to interact with gebeya Saas using natural language queries through multiple channels.
This project enables users to interact with gebeya Saas using natural language queries through multiple channels. I was able to work on - NLU & NER: Initially built core components using Rasa for intent classification and named entity recognition before transitioning to more advanced, LLM-native function calling to handle complex user queries. Multi-Channel Integration: Ensured robust, scalable communication across multiple platforms (Telegram, WhatsApp, Web UI) by engineering a custom webhook-based backend to manage connections between the platforms and agentic workflows. System Observability: Integrated Langfuse for detailed tracing, logging, and Redis caching of all AI interactions, capturing user sessions, API costs, requests, and model responses to optimize performance and cost.
Autonomous Video Subtitling System
An R&D project for an autonomous video subtitling system that uses Generative AI and RAG to subtitle videos.
Led an experimental project to build an autonomous system for subtitling and dubbing videos from English to Amharic. Achieved high accuracy in subtitling, while identifying and documenting key challenges in maintaining speaker tone, gender, and speed for AI-powered dubbing.