🚀 Theme 1: Open Model Evaluation – AI Benchmark & Scorecard

For Hacktoberfest Hack Day, our team built a comprehensive benchmarking and evaluation framework comparing state-of-the-art open-weight AI models (Gemma 4 31B IT, Gemma 4 26B A4B MoE, gpt-oss-120b, and gpt-oss-20b) across 25 standardized prompts spanning Reasoning & Math, Factual Accuracy, Instruction Following, Coding, and Indian Languages.

                Key Highlights & Deliverables:

• Empirical Scorecard: Mapped 100+ individual test scores across platforms (Google AI Studio & Groq Playground) to evaluate overall accuracy and latency trade-offs.

-MoE vs. Dense Analysis:Uncovered real-world performance patterns, demonstrating how Mixture of Experts (MoE) models trade off niche regional trivia (e.g., Kannada state bird trivia) for lightning-fast inference speeds.

Production-Ready Artifacts:Fully documented fact sheets, automated Google Sheets scorecard, and an interactive side-by-side live testing suite ready for live judge demonstrations.