PSL Calculator

This submission for the Hacktoberfest Open-Source AI Challenge Week 3, titled "PSL Calculator," showcases an architecture built on an open and flexible stack, optimized for speed and transparency. The AI & Inference component relies on open-weight models like Gemma 4, integrated via lightweight local harnesses and custom Python vision pipelines.

The Backend & Processing is constructed in Python, managing real-time state management, custom prompt structuring, and multi-agent analytical logic, along with seamless HEIC image parsing. The Frontend is built using Streamlit, offering a clean, reactive, and minimalist UI where users can upload images and review detailed evaluation breakdowns.

Open innovation played a crucial role in developing this tool, as closed, proprietary APIs impose rigid constraints that hinder experimentation. Unrestricted Customization enabled the team to fine-tune system prompts, adjust scoring rubrics, and inspect raw inference outputs without facing arbitrary rate limits or hidden payload filters.

Cost & Performance Control were also significant advantages, as running the pipeline with open tools allowed for local optimization, eliminating unnecessary network overhead and ensuring dependable, rapid execution. Community Extensibility ensured that anyone could audit, fork, and build upon the codebase, maintaining transparency and adaptability.

The streamlined local workflow allowed the team, consisting of Annukul Ghale and Mustafe Musse, to achieve a fully deployed, interactive web app in record time. This approach saved them precious time, bypassing the need to configure rigid cloud infrastructure, grapple with proprietary SDK wrappers, or manually convert iPhone HEIC photos into JPEGs.