Birehan Zewdie's CV

New York, USA (Remote)

• ›7% revenue increase from an AI floor-price prediction system (LightGBM) that optimizes pricing per ad request for DSPs.
• ›86%+ accuracy on a client revenue-forecasting system built from scraped domain data, PageSpeed and WHOIS signals, historical performance, and LLM-generated features.

London, UK (Remote)

• ›~16% KPI lift from an automated pacing algorithm that pairs predictive modeling with real-time bid adjustments.
• ›~10% engagement lift and ~5% CTR/CPA improvement from a creative scoring system (historical performance + LLMs) that ranks variants before spend.
• ›Cut creative ideation time roughly in half with an LLM pipeline that generates on-brand ad concepts.
• ›~15% conversion lift and ~10% margin lift from an inventory scoring model (DCN v2 + learning-to-rank) that tunes bids across publishers, formats, and devices.

Addis Ababa, Ethiopia

• ›Led a cross-functional team of 18 to launch HakimHub, a first-of-its-kind AI-powered medical recommendation platform, coordinating planning and delivery through Jira.
• ›Fine-tuned a Llama 3-based chatbot using AI agents for personalized symptom assessments and follow-ups, integrated with tooling to query doctors and hospitals.

BSc in Software Engineering (AI Stream)

OOP, Databases, Software Development, Machine Learning, NLP, Reinforcement Learning.

Data Science · ML · Generative AI · Web3 (Distinction, Valedictorian)

Software Engineering Program (backed by Google)

Solved 1,000+ algorithm problems (850+ LeetCode, 200+ Codeforces).