[Paper Reading]: R2Code: A Self-Reflective LLM Framework for Requirements-to-Code Traceability

📄 Weekly Paper Reading Session

Join SupportVectors AI Training Lab for another deep-dive into cutting-edge AI research! This week's session walks through and discusses the paper:

🔬 R2Code: A Self-Reflective LLM Framework for Requirements-to-Code Traceability

Speaker: Krishnan Ramaswamy

---

🧠 About the Paper

Accurate requirement-to-code traceability is a cornerstone of effective software maintenance — yet current IR- and embedding-based approaches lean heavily on lexical similarity, frequently producing incomplete or inconsistent trace links across different projects and programming languages, while also incurring significant costs from long-context retrieval and prompting.

This paper introduces R2Code, an LLM-based semantic traceability framework engineered to boost trace link accuracy while simultaneously cutting inference costs. The framework is built around three core components:

• 🔗 Bidirectional Alignment Network (BAN) — A decomposition-enhanced module that aligns four-layer requirement semantics with corresponding code structures, enabling robust cross-level semantic matching
• ✅ Self-Reflective Consistency Verification (SRCV) — An explanation-guided consistency-checking module that calibrates the reliability of trace links
• ⚡ Dynamic Context-Adaptive Retrieval (DCAR) — A mechanism that adjusts retrieval granularity and filters contexts via semantic-overlap weighting for efficient context utilization

📊 Key Results

Experiments conducted across five public datasets spanning multiple domains and two programming languages show that R2Code consistently surpasses the strongest baselines — achieving an average F1 gain of 7.4% while slashing token consumption by up to 41.7% through adaptive context control.

---

📍 Attendance Options

This session is open to everyone — attend in person at SupportVectors in Fremont, CA (conveniently located near Tesla and accessible via road and BART), or join remotely via Zoom.

---

🏫 About SupportVectors

SupportVectors AI Training Lab delivers high-quality, hands-on, industry-targeted AI training designed to keep professionals at the forefront of the field. Everyone is welcome — whether you're a seasoned ML engineer or just getting started with AI research! 🚀