The Software Developers Association at ASU

?? **Thrilled to Announce! Unbiasly Wins 2nd Prize at Hack SoDA 2024!** ?? We built our AI-powered Chrome extension to detect and correct biases in real-time across both written and spoken language, creating a tool that promotes inclusive online communication ??. Unbiasly was inspired by the need to address unconscious bias in everyday language. Features include giving alternative unbiased suggestions using Gemini's API, scanning the webpage, and using speech-to-text to give bias metrics for different areas using Flair ML model. Our frontend is built in JavaScript, with a Python-powered backend that processes inputs and delivers quick, helpful suggestions. Winning 2nd place at Hack SoDA by The Software Developers Association at ASU is an incredible milestone ?? for our team of four: Sai Mahesh Nomula, Nandhini Gopireddy, Abhijna Venkatesh Maiya and myself. We’re proud of Unbiasly’s potential to make a meaningful impact. This journey reinforced our commitment to tech for social good, emphasizing the importance of user experience and accessibility. https://lnkd.in/dxPGtAe9

?? Wrapped up an incredible journey at HackSoDA, a 24-hour hackathon organized by The Software Developers Association at ASU at Arizona State University! While I didn’t take home a win, the experience was absolutely worth it, packed with new learnings, teamwork, and inspiration. ???? From brainstorming with brilliant minds ?? to navigating challenges under time pressure ?, HackSoDA pushed me to think creatively and refine my problem-solving skills. I gained hands-on experience with new technologies, expanded my coding toolkit ???, and connected with like-minded peers and mentors who are just as passionate about tech innovation ??. The highlight? Witnessing the diversity of ideas and the creative solutions everyone brought to the table ???. It was a powerful reminder of the impact technology can have on real-world problems, and I’m more motivated than ever to keep building, learning, and growing ??. Thank you to the organizers, mentors, and fellow participants for an unforgettable experience. Can’t wait to see what’s next! ???? #HackSoDA #SoDA #Hackathon #StudentDevelopers #TechInnovation #GrowthMindset #ASU

Model to Detect Bias in Vietnamese and Hindi Sentences ???? | Winner of HackSoDA 2024 Data Science for Social Good Track ??? ?? Did you know that word embeddings — the representation of words in vector space — fundamental to training LLMs like ChatGPT ?? — can carry social biases? To illustrate, I did a simple visualization of the word embeddings of "Men", "Programmer", "Women", and "Homemaker" in English using Word2Vec ??. I discovered that the distance between "Men" and "Programmer" is almost identical to the distance between "Women" and "Homemaker" — a bias that influences how LLMs interpret words. ??? Solution: I decided to build an algorithm to detect social bias, starting with gender bias, in Vietnamese and Hindi within 24 hours. My goal was not only to mitigate bias in LLM training corpus but also to boost the representation of diverse languages in NLP research ??. ?? Algorithm Overview This model leverages cosine similarity, principal component analysis (PCA), and concepts from the following foundational papers: ?? Estimating Gender Bias in Sentence Embeddings by Tommaso Dolci ?? Man is to Computer Programmer as Woman is to Homemaker? Debiasing Word Embeddings by Tolga Bolukbasi Since these methods were initially applied in English, I reverse-engineered the algorithms for Vietnamese and Hindi, achieving an impressive 80% accuracy on a dataset of 40 sentences across these languages ??. ?? Shout out to my teammate Dheeraj Kallakuri for his exceptional support! We met during the team-up section, and he did an amazing job deploying the frontend and sourcing an effective pre-trained embedding model for Hindi ??. Special thanks to The Software Developers Association at ASU for hosting this fantastic hackathon ??! ?? Next Steps This project won’t stop at the hackathon! I aim to extend it by testing with Vietnamese language models and eventually releasing a Python package ??. This will enable users to test bias in their preferred language, using their own language models and bias measures ????. If you’re interested, please star the GitHub repo ? linked below as I will announce more development on this! ?? Demo: https://lnkd.in/gnF9ZHQU ?? GitHub: https://lnkd.in/geUUR5-v ?? Devpost: https://lnkd.in/gCKvk3uQ #hackathon #ASU #generativeAI #GenAI #biasmodel #biasdetection #transformer #tokenization #regionallanguages #soda #datascience #embeddings #NLP #Python #languageBias #streamlit #flask