Pedro Orvalho

(4/4) 🔥 Join us at AAAI 2025! Catch our poster on Sat, March 1 at 12:30 🎉

📄 Read the full paper here: arxiv.org/pdf/2502.07786

(3/4) 🔁 If the LLM makes mistakes, we feed it counterexamples, refining its fix step by step—just like teaching it through debugging! 🎯

✅ Tested on 1,431 student programs, our method repairs more code than standard repair tools! 📊

(2/4) 🔧 Our hybrid approach: MaxSAT-based fault localization + LLM-guided synthesis 🔄
1️⃣ Detect buggy code 🐞
2️⃣ Remove errors, creating a program sketch ✂️
3️⃣ Let the LLM fill in the gaps 🏗️
4️⃣ Validate fixes with a test suite ✅

🚀 LLMs + Formal Methods = Smarter Program Repair? Our paper was just accepted at AAAI 2025! 🎉

🔍 Formal Methods find bugs but struggle with fixes. 🤖 LLMs repair code but over-edit. What if we combined their strengths? 🧵👇

(5/5) 🚀 Ready to transform programming education? Read our preprint on GitSEED here: arxiv.org/abs/2409.07362 📖

👩‍💻 Try it for yourself and see the magic in action: gitlab.inesc-id.pt/u020557/GitS... 🌟

👀 Catch our presentation tomorrow at SIGCSE Virtual 2024: sigcsevirtual.acm.org

#SIGCSE #CSforALL

(4/5) 👩‍🎓💬 Results? 91.8% of students say GitSEED boosted comprehension! From submission to feedback to revision, this centralized platform made learning smoother & more engaging. 🔥

(3/5) 💡 GitSEED is easily customizable! Faculty can tailor it to integrate tools like:
Hints: Fault localization/static analysis guide students to errors.
Valgrind: Debug memory issues effortlessly.
Lizard: Analyze code complexity & optimize it.
Students love these features—why not add your own? 🚀

(2/5) 📊 Dashboards are a game-changer! GitSEED tracks:
✅ Successful/failed tests
✅ Submission counts
✅ Days since start of assignments

CS students see their progress and their peers’, fostering motivation & accountability! 🚀

🎓⚙️ Meet GitSEED, a revolutionary tool for programming education accepted at sigcsevirtual.acm.org! Labs, projects, dashboards & personalized feedback—all on @gitlab.com. Let’s dive into how it transforms learning. 🧵👇