This is a submission for the Hacktoberfest Open-Source AI Challenge Week 3: Slow and Steady
What I Built
Demo
https://psl-calculator-mlh-hack-5dnqeov6yyaqpqkqq9eezj.streamlit.app
Code
https://github.com/Vedu56/PSL-Calculator-MLH-Hack
How WE Built It
Our architecture relies on an open and flexible stack optimized for speed and transparency:
AI & Inference: Powered by open-weight models (such as Gemma 4 workflows) integrated via lightweight local harnesses and custom Python vision pipelines.
Backend & Processing: Built in Python, handling real-time state management, custom prompt structuring, and multi-agent analytical logic alongside seamless HEIC image parsing.
Frontend: Built using Streamlit to provide a clean, reactive, and minimalist UI where users can instantly upload images and inspect detailed evaluation breakdowns.
Why Does Open Innovation Matter?
Open innovation was crucial for building this tool because closed, proprietary APIs introduce rigid constraints that stifle experimentation:
Unrestricted Customization: Having access to open-weight vision models allowed us to fine-tune system prompts, adjust underlying scoring rubrics, and inspect raw inference outputs without hitting arbitrary rate limits or hidden payload filters.
Cost & Performance Control: Running our pipeline with open tools enabled local optimization, removing unnecessary network overhead and ensuring reliable, lightning-fast execution.
Community Extensibility: An open architecture means anyone can audit, fork, and build directly on top of our codebase, keeping the project transparent and endlessly adaptable.
My Agent Session
The Lazy Payoff
Instead of spending days configuring rigid cloud infrastructure, wrestling with proprietary SDK wrappers, or manually converting iPhone HEIC photos into JPEGs, our streamlined local workflow allowed us to go from concept to a fully deployed, interactive web app in record time. It saved us dozens of hours of trial-and-error this week alone!
Prize Categories
Best open source and best use of Gemma API.
Teammates were Annukul Ghale and Mustafe Musse
Top comments (0)