- 🔬 I do research on LLM alignment/RLHF, multi-agent systems, and AI red-teaming/security — previously at Adobe MDSR, Prem AI, Repello AI, Boston University, and Lossfunk.
- 📝 My work on preference optimization (IPO) and reasoning alignment (ACL 2026 Findings) has been published at ACL Main Conference / Findings and AAAI workshops.
- 🛠️ Currently building agentic pipelines, RL environments for LLM agents, and test-time adaptation methods — see pinned repos below.
- 💬 Ask me about LLMs, RLHF/DPO/GRPO, multi-agent systems, or AI security/red-teaming.
- 📫 Reach me at ayushsingh73920@gmail.com
- CATPO: Critique-Augmented Tree Policy Optimization — arXiv Preprint, 2026 — Paper
- Thinking About Thinking: Evaluating Reasoning in Post-Trained LLMs — ACL 2026 Findings — Paper
- IPO: Your Language Model is Secretly a Preference Classifier — ACL 2025 Main Conference — Paper
- LoRA-Mini: Adaptation Matrices Decomposition — AAAI CoLoRAI Workshop — Paper
- Adaptive Urban Planning: A Hybrid Framework — AAAI AI4UP Workshop — Paper
- 🔭 RAVEN — LangGraph hypothesis-validation copilot generating audit-ready financial research reports.
- 🏗️ StructuralDesignEnv — OpenEnv RL environment where LLM agents design Eurocode-3-compliant steel frames.
- 🌊 RealPDE-LTTTA — Long-term test-time adaptation for streaming real-world PIV flow forecasting (NeurIPS 2026 competition).
- 🛡️ JED Red-Team — Multi-step tool-attack search against guardrailed LLM agents (Kaggle).
