The Future of AI Agents: Results & Prospects

A review presents the results of the OllamaModelTester platform after twelve online sessions with an open door to decide where the platform goes next. During the meeting with OllamaModelTester platform four AI models were tested in real time. Faithfulness scores were displayed side by side with hallucination detection and the dashboard visualization. Chapters: 0:00 - Event Introduction 1:40 - Poll 1 - Problems with AI Agents 8:12 - Present Solution - OllamaModelTester 16:29 - Live Demonstration 28:25 - Review Results from Test 48:35 - Poll 2 - Nominate the AI Model Winner 55:37 - Summary of Presentation 56:58 - Poll 3 - The Future of AI Model Comparison 1:06:05 - Event Ending 1:09:09 - Special Thanks Technical Architecture: • Full-stack application: FastAPI backend, React frontend • Deployed on GCP VM instance • Containerized with Docker, orchestrated with Kubernetes (Minikube) • Automated CI/CD pipeline via GitHub Actions • Hardware metadata awareness built in Tested AI Models: • Tinyllama (Singapore University of Technology and Design) • Smollm2 (Hugging Face) • Gemma2 (Google) • Llama3.2 (Meta) Live Polls: • Poll 1: Problems with AI Agents. Have you found incorrect AI-generated content? And where? • Poll 2: Nominate the AI Model Winner • Poll 3: The Future of AI Model Comparison. How should AI trust be measured? The AI Model Winner: After equality in the votes and a discussion, the champion is Gemma2 (Google). Congratulations to the winners!