Evidence & grounding
Evidence-grounded language models
Using retrieval, knowledge graphs, and collaborating agents to connect language generation with trustworthy sources.
Ph.D. student · Computer Science & Engineering
Anirban
Saha Anik.
I build language systems that are reliable, useful, and human-centered.
I am a Ph.D. student at the University of North Texas, working on trustworthy language models for health and crisis communication. My research connects reliable evidence, adaptive explanations, and rigorous evaluation.
Advised by Dr. Lingzi Hong · Human-Centered Computing Lab
Open to 2027 research internships in Applied Science, NLP, and Machine Learning.
Get in touch01 / RESEARCH
What I work on.
Developing AI that not only generates fluent answers, but communicates evidence with the right context and detail.
Explore all researchPeople & interaction
Adaptive human–AI interaction
Designing systems that know when to ask, answer, and adapt explanations to the needs of different readers.
Reliability & evaluation
Evaluation for high-stakes NLP
Building benchmarks, metrics, and human studies for factuality, consistency, usefulness, and situational awareness.
RESEARCH INTERESTS
Large Language Models · Retrieval-Augmented Generation · Knowledge Graphs · Human-Centered AI · Multi-Agent Systems
02 / SELECTED WORK
Publications.
Ask or Answer: A Decision Framework for Multi-Turn Health Misinformation Intervention
Deciding when an intervention should ask for clarification or answer directly.
Multi-Agent Retrieval-Augmented Framework for Evidence-Based Counterspeech Against Health Misinformation
An evidence-grounded multi-agent framework for health misinformation correction.
Runner-up · SIG-SM Social Media Research Competition
Speaking at the Right Level: Literacy-Controlled Counterspeech Generation with RAG-RL
Adapting evidence-grounded corrections to different health-literacy levels.
Finalist · SIG-SM Social Media Research Competition
A Dynamic Fusion Model for Consistent Crisis Response
Producing more consistent, professional, and actionable crisis responses.
04 / CONNECT
Let's talk research.
Interested in trustworthy NLP, human-centered AI, or potential collaborations? I would be happy to connect.