AI ATL

Thank you all for coming out to AI ATL 2024 — Atlanta’s AI hackathon at Georgia Tech, hosted by Startup Exchange and AI @ Georgia Tech and brought to you by Google Cloud! 🚀 This weekend was filled with creativity and innovation, thanks to your hard work and dedication! We hope the workshops from Google Cloud, Anthropic, Microsoft, and NLX were engaging and informative! Our team also wanted to extend our heartfelt congratulations to the winners of this year's hackathon. You should be incredibly proud of the time and effort you put into your projects. A big shoutout to the track winners of AI ATL 2024: Overall: KayEcho.AI 🏆 Beginner: Dancely 🏅 Find the rest of the challenge winners here: https://lnkd.in/gyZgWV_R We are super proud of the time and effort all of you put into this weekend and we are here to support you! Check out the rest of the projects on the Devpost here: https://lnkd.in/gq7QQvNj We hope you had as much fun as we did and felt inspired by all the talent in the room. We can’t wait to see where your projects go from here.

🎬 Thrilled to share our AI ATL Hackathon project through video! Introducing DayCanvas - Where Your Journal Comes to Life! 🌟 Watch the full demo here: https://lnkd.in/gYC8BMDE At Georgia Institute of Technology's AI ATL Hackathon, Leonardo Lujan and I tackled a challenge many face: the struggle to process and understand our daily emotions. Our solution? DayCanvas, a platform that transforms journal entries into visual stories. 🛠️ Technical Deep Dive: Emotional Analysis: Leveraged Anthropic's Claude 3.5 Sonnet model to: - Extract and analyze emotional patterns from journal entries - Identify core emotional themes - Generate emotion-focused narrative summaries - Provide contextual understanding of emotional states Visual Generation: Implemented Google's Imagen 3 (Special Thanks to Paige Bailey for giving us access to this) for: - Creating emotion-centric visual narratives - Generating scenes that reflect emotional intensity - Maintaining visual consistency across story episodes - Translating emotional context into visual metaphors Integration Framework: - Real-time processing of journal entries - Seamless pipeline from text analysis to image generation - User-friendly interface for journal input and story viewing 🎯 Key Features: - Transform written emotions into vivid visual narratives - AI-powered image generation matching your emotional journey. - Each day becomes an episode in your personal story - Track emotional patterns through data-driven insights 💫 Impact Highlights: - Makes emotional reflection more engaging and accessible - Helps users process complex feelings through creative visualization - Provides data-driven insights into emotional patterns - Creates a safe space for emotional expression and growth - This journey wouldn't have been possible without: - The incredible mentorship at Georgia Tech My amazing teammate Leonardo Lujan The supportive AI ATL Hackathon community Access to cutting-edge AI models from Anthropic and Google Check out the video to see how we're pushing the boundaries of AI for emotional wellness! Would love to hear your thoughts and feedback. 🙌 #AIHackathon #MentalHealth #Innovation #AI #GeorgiaTech #PersonalDevelopment #EmotionalWellness #TechForGood #AnthropicAI #GoogleAI

I enjoyed developing ClassMage with  Theo Halpern, Varindra Seegobin, and Nedu Genas at the AI ATL hackathon! ClassMage is your new AI-powered study buddy for professors! Students can use ChatGPT for classwork. However, conversational AIs often hallucinate or provide irrelevant information. Using Google Gemini with RAG with Vite + React on the front-end, professors can create a class, upload course materials such as slideshows, and create a Chatbot link for that class. When a student uses that link and asks ClassMage questions, ClassMage will scan and weigh the relevance of the uploaded documents to provide answers to the student's question only from that source. Then, it will display an answer to the student's question and the document it referenced! In this hackathon, I learned some Meta business practices from the Meta SWE intern Theo Halpern, such as what useEffect() actually means and how to "componentize" my code. As I was responsible for developing the front end, I learned TailwindCSS! Although we didn't win, I learned how to use Google Cloud services in 24 hours!

🏆 Thrilled to share our project from AI ATL 2024 Hackathon: "Are You Sure About That?" - a game-like platform that puts humans against Claude 3.5 in detecting flawed AI reasoning. Built natively in Python using Streamlit. Deployed with DigitalOcean, Cloudflare, and Docker, Inc. With the rise of LLMs, we wanted to tackle a crucial question: Can AI models spot unfaithful reasoning in other AI outputs? And how do humans compare? Our submission to the Alignment track explores this critical aspect of AI safety. The platform tests both humans and AI on their ability to identify flawed logic in AI responses, creating a unique benchmark for reasoning capabilities. 🎯 Key Features: • Interactive Chain-of-Thought evaluation • Real-time AI vs Human performance tracking • Human feedback collection for model improvement • Comprehensive scoring system Video: https://lnkd.in/ex_tKhRy Github: https://lnkd.in/ej3WM62z Devpost: https://lnkd.in/e3NV3gYF 🔍 Try it yourself: https://lnkd.in/eMCgp6Dq 📚 Read more about our methodology and findings in our alignment track paper: https://lnkd.in/e6PaiCSp Thanks to AI ATL for organizing this amazing event! Looking forward to more opportunities to contribute to AI safety research.

Take a moment to 𝗿𝗲𝗺𝗶𝗻𝗱 𝘆𝗼𝘂𝗿𝘀𝗲𝗹𝗳 𝗼𝗳 𝘁𝗵𝗲 𝗽𝗼𝘄𝗲𝗿 𝗼𝗳 𝗮 𝗳𝗿𝗲𝘀𝗵 𝘀𝘁𝗮𝗿𝘁. 🌟 This weekend, I had the privilege of being a Mentor and Judge at the AI ATL hackathon at Georgia Institute of Technology, where over 100 teams came together to turn ideas into reality. In just 36 hours, these students poured everything they had into their projects, sacrificing sleep for the chance to create something new. Watching them took me back to my own college days at Northeastern University when passion and curiosity fueled those late nights working on something I truly believed in. It’s easy, as professionals, to forget the thrill of doing things for the very first time—the excitement, the eagerness to learn, and the drive to keep going no matter what. Being at this event reminded me how powerful that energy can be. For all my peers reading this: 𝗧𝗮𝗸𝗲 𝗮 𝗺𝗼𝗺𝗲𝗻𝘁 𝘁𝗼 𝗿𝗲𝗰𝗼𝗻𝗻𝗲𝗰𝘁 𝘄𝗶𝘁𝗵 𝘁𝗵𝗮𝘁 𝗲𝗮𝗿𝗹𝘆 𝗱𝗿𝗶𝘃𝗲, 𝗰𝘂𝗿𝗶𝗼𝘀𝗶𝘁𝘆, 𝗮𝗻𝗱 𝗱𝗲𝘀𝗶𝗿𝗲 𝘁𝗼 𝗹𝗲𝗮𝗿𝗻. Here are a few takeaways I gained from working with the next generation of innovators:  ✅ 𝗘𝗺𝗯𝗿𝗮𝗰𝗲 𝗰𝘂𝗿𝗶𝗼𝘀𝗶𝘁𝘆—it’s what drives us to grow beyond our comfort zones.  ✅ 𝗗𝗼𝗻’𝘁 𝘄𝗮𝗶𝘁 𝗳𝗼𝗿 𝗽𝗲𝗿𝗳𝗲𝗰𝘁𝗶𝗼𝗻—take action, learn as you go, and trust the process.  ✅ 𝗥𝗲𝗺𝗲𝗺𝗯𝗲𝗿 𝘁𝗵𝗲 𝗽𝗼𝘄𝗲𝗿 𝗼𝗳 𝗳𝗶𝗿𝘀𝘁 𝘁𝗶𝗺𝗲𝘀—the thrill of doing something new can fuel the toughest days. Rediscovering that “college self” now and then can add a lot of meaning and excitement back into our careers. 𝗟𝗲𝘁’𝘀 𝗸𝗲𝗲𝗽 𝗽𝘂𝘀𝗵𝗶𝗻𝗴, 𝗹𝗲𝗮𝗿𝗻𝗶𝗻𝗴, 𝗮𝗻𝗱 𝗴𝗿𝗼𝘄𝗶𝗻𝗴. #HackathonLife #AIATL #KeepLearning P.S.: The link to all the projects is in the comments below. Do check them out!

This past weekend, I participated in the AI ATL Hackathon, Atlanta's First AI Hackathon, and the first hackathon I have ever attended. Over the 36-hour hackathon, my teammate Daniel Achacon and I developed Swida, an AI-powered lung cancer assistant. Swida uses Image Transformers, Grad-CAMs, and Google Gemini in order to provide a full diagnostic explanation given a lung CT scan. How it works: First, the user inputs a lung CT scan and clicks on upload. This then calls the backend API to classify the CT scan into whether or not it is cancerous, using a fine-tuned Data Efficient Image Transformer from HuggingFace. Next, it will run the Grad-CAM algorithm, which was trained using a resnet50 CNN and the pytorch_gradcam library, to generate a visual heatmap over the CT scan to show exactly where the AI is looking when it comes to its diagnosis. Finally, it calls Google-Gemini-1.5-pro & flash in order to explain the diagnosis, provide potential treatment options for the CT scan, analyze the Grad-CAM heatmap, and explain the location-specific challenges of the cancer. Finally, it uses Gemini & the PubMed API to pull cutting-edge research papers related to the diagnosis. We used React for the frontend & Flask for the backend. Conclusion: This was such an amazing learning experience, and I had a lot of fun over this hackathon. I wanted to give a special thank you to my teammate Daniel, who was crucial to the development of Swida and helped make the sleepless nights more bearable, and also a special thank you to Startup Exchange, Artificial Intelligence @ Georgia Tech (AI@GT), and the sponsors Google, Anthropic, Microsoft, and more for the free merch and technology they shared over the weekend. This was a great start on my hackathon journey, and I am excited to attend even more in the future! Here are some links to learn more about how we made Swida: Devpost submission: https://lnkd.in/gPNfDUM9 Github: https://lnkd.in/gtkhqbpQ

Had an amazing time hacking at the AI ATL Hackathon this weekend! My team built justis[.]ai, an agentic legal assistance system built using LangChain's LangGraph to provide legal assistance to people who don't have access to lawyers. The LLM agents powered by both Anthropic's Claude and Google's Gemini work in tandem to understand the user's grievances, make calls to legal APIs to get close-to-ground truth information, and fill legal documents on the user's behalf. We built this in 36 hours with no sleep and lots of caffeine! While it was fascinating to see it do such a great job at autonomously grabbing and filling legal documents, we did face a lot of the challenges that come with providing LLMs autonomy to take actions. Nevertheless, I think the future is agentic and I'm all for it. Devpost: https://lnkd.in/gvaMMwYK Repo: https://lnkd.in/gzB-P4iP

Grateful to have been part of the team at the AI ATL Hackathon. It was a great learning experience working with Hemantha Krishna Bharadwaj and Deepak Srinivas to create a multi-agentic legal assistance system using LangChain's LangGraph, powered by Anthropic's Claude and Google's Gemini, to help make legal aid more accessible. Special thanks to Hemantha Krishna Bharadwaj for giving deep drive through the new domain of LLMs, RAGs, LLM agents and ReAct systems.

In just 36 hours, my team and I built justis[.]ai, an agentic legal assistance system aimed at helping individuals without access to legal support. The platform leverages LangChain's LangGraph to orchestrate LLM agents that collaborate to understand user's issues, retrieve real-time legal information, and even assist in drafting documents—all designed to make legal assistance more accessible and efficient. I had the privilege of helping build the front end, creating an intuitive interface that bridges complex backend functionality with a seamless user experience. From late-night brainstorming sessions to debugging and a copious amount of caffeine, this experience was a true test of teamwork and resilience. I'm truly thankful for the chance to work with an amazing team to build some cool software and explore the world of agentic systems on a deeper level! Github: https://lnkd.in/e3cEUKPx Hemantha Krishna Bharadwaj Viren Bhosale AI ATL Anthropic Google LangChain Alex Albert

AI ATL Project Demo: DayCanvas - Where your Journal comes to life 🔥. Demo Link: https://lnkd.in/eXTsd9SJ This past weekend I attended the annual AI ATL hackathon at Georgia Institute of Technology. My teammate Sahil Mangotra and I sought a way to process and understand our daily emotions, and came up with DayCanvas. DayCanvas turns user uploaded journal entries into visual stories, to help users view their experiences with the help of AI to improve their emotional maturity and wellness. After an insightful presentation from Alex Albert we familiarized ourselves with Anthropic’s Claude 3.5 Sonnet model, to identify emotional themes in the user’s journal entries. We also used Claude to provide contextual understanding of various emotional states from the user’s responses. Utilizing Google’s Imagen 3, we inputted our Claude generated emotional themes and context to generate images for each part of the journal entry. Shout out to Paige Bailey for an insightful workshop and access to work with this tool! Designing the user interface in NextJS allowed us to polish our project, opting for a minimal interface that allows our users to immerse themselves in the colors of the generated images. Not only did I get the chance to work on a project so personal to me, but I also learned how to work with APIs and AI tools for the first time. I walked away from the hackathon feeling excited to learn more about integrating AI tools within the projects I make. I want to give a large thanks to the AI ATL team, my teammate Sahil, and all the professionals who made this event possible 🔥