Research
My research investigates conversational and agentic AI systems, multi-turn intent classification, and retrieval-augmented NLP in production environments, alongside foundational work in quantum-classical hybrid learning.
Research themes
- →Agentic AI architectures, multi-agent coordination, and tool-use reliability
- →Multi-turn intent classification and conversational dialogue generation
- →Retrieval-augmented generation (RAG) for low-resource and multilingual deployments
- →Quantum-classical hybrid neural networks and adaptive learning systems
Selected publications
Addresses the challenge of deploying accurate multi-turn intent classification in production conversational AI systems where latency and scalability conflict with model size. Introduces Symbol Tuning and C-LARA (Consistency-aware, Linguistics Adaptive Retrieval Augmentation), achieving a 5.09% accuracy improvement while reducing annotation costs by 40% in multilingual industrial deployments.
Introduces Chain-of-Intent, integrating Hidden Markov Models (HMMs) with LLMs to synthesize domain-specific, context-aware multilingual dialogues via self-play, alongside MINT-CL for multi-task contrastive learning and the public MINT-E multilingual e-commerce dialogue corpus.
Presents a Linguistic-Adaptive Retrieval-Augmentation framework enhancing multi-turn intent classification across six languages. Combines a fine-tuned compact model with adaptive retrieval over historical dialogues, achieving +3.67% average accuracy over single-turn baselines in production.
Proposes a hybrid quantum-classical convolutional neural network (QCCNN) compatible with NISQ devices, pairing quantum circuit feature extraction with classical deep learning and automatic hybrid gradient computation.
Open Science
I am committed to Open Science: publishing open-access preprints, releasing curated industrial benchmarks (such as MINT-E), and documenting reproducible architectures that bridge academic rigor and production constraints.
View full publication record (900+ citations) on Google Scholar.