I develop intelligent interactive AI systems as experimental platforms to study how humans understand, trust, and adapt to AI under uncertainty.
My work lies at the intersection of:
- 🤖 Human-AI Interaction — Understanding how people reason about, trust, and adapt to AI systems.
- 🧠 Computational Cognitive Modeling — Formalizing psychological theory to model human reasoning under uncertainty.
- ⚙️ Adaptive Intelligent Systems — Engineering AI systems that respond to and align with human cognition.
- 🔍 Explainable AI — Informing AI design to improve transparency, interpretability, and user confidence.
I pursue a cognition-driven, closed-loop research program to develop intelligent systems that adapt to human reasoning under uncertainty.
| 🧠 Question | 🏗️ Build | 🧪 Study | 🔢 Model | 🔄 Adapt |
|---|---|---|---|---|
| Formulate cognitive and interaction questions about how humans interpret, trust, and update beliefs about AI behavior. | Develop controlled, interactive AI systems as experimental platforms. | Conduct behavioral experiments to measure trust dynamics, belief updating, and mental model formation over time. | Construct computational models that infer latent cognitive states such as beliefs and uncertainty. | Embed these models into AI systems that dynamically adjust explanations, autonomy, and decision policies—enabling human–AI co-adaptation. |
This cycle moves from understanding human cognition to engineering human-adaptive intelligent systems grounded in empirical and computational insights.
Belief-based computational modeling of human social reasoning in AI interaction. Technical Contribution: Developed a probabilistic framework using Pyro to simulate how humans infer social intent from agent behavior. Modeled belief updating dynamics and trust calibration mechanisms. Research Impact: Introduced the Social Intelligence Quotient (SIQ) to quantify human-agent alignment. Provides a foundation for socially adaptive AI systems. Stack:
|
Interactive human-in-the-loop auditing system for fairness in NLP decision systems. Technical Contribution: Integrated DistilBERT with SHAP attribution and counterfactual explanation generation. Built an interactive Gradio interface for real-time bias exploration. Research Impact: Enables users to understand, audit, and correct algorithmic bias. Demonstrates how explainability affects trust calibration. Stack:
|
|
Real-time estimation of user cognitive load from natural interaction traces. Technical Contribution: Designed an end-to-end behavioral inference pipeline that models cognitive load using interpretable interaction features. Achieved F1 = 0.82 under participant-independent evaluation using machine learning. Research Contribution: Conducted a controlled user study (N = 25) to collect high-resolution behavioral telemetry and NASA-TLX workload ground truth for validating cognitive load inference. Stack:
|
Transformer-based framework for antimicrobial peptide discovery. Technical Contribution: Built deep learning pipelines for sequence modeling. Improved predictive accuracy beyond prior baseline methods. Research Impact: Supports AI-driven drug discovery for WHO priority pathogens. Stack: Broader Significance: Demonstrates my ability to design high-performance transformer-based systems, informing my work on human-adaptive intelligent systems.
|
| 📄 Publication | 📚 Venue | 🔗 Links |
|---|---|---|
| UNET-Based Segmentation for Diabetic Macular Edema Detection in OCT Images | ICCIS 2025, Springer LNNS |   |
Proposed and implemented the Stack:
|
A theory-driven dataset for modeling human trust and belief updating in human–AI interaction. Grounded in computational cognitive science principles. Stack: |
Human-in-the-loop LLM adaptation framework with LoRA/PEFT and stability–plasticity evaluation for continually adapting language models to human feedback. Stack: |
