Essam Mansour

Manually creating dialogue benchmarks costs thousands, but our Chatty-Gen automates it with #RAG & LLM-as-a-Judge to deliver human-level quality for <$1! 🚀 Our modular approach could automate different synthetic benchmarks & enable LLMs as domain experts for data science & security. More to come 🏃🏻‍♂️‍➡️!

🎉 Our paper Chatty-Gen is accepted at SIGMOD 2025! 🚀 Huge thanks to my students Reham Omar & Omij Mangukiya! Our approach prevents costly errors & works across LLMs like GPT-4, Gemini, Llama-3 & Mistral. We combined Llama-3 & CodeLlama2 to match GPT-4o! Read more arxiv.org/abs/2501.09928 #SIGMOD2025