Human–AI collaboration
Designing a four-arm experiment on collective action
Co-developing a working paper and online study of how AI interpretation and coordination support affect decisions in shared supply chains.
Computer Science, NYU Abu Dhabi
Ashmit Mukherjee
I study when AI systems improve model performance—and how they change human decisions.
My work joins parameter-efficient model adaptation with human-AI experimentation. I build datasets and training pipelines, design controlled interventions, and test whether apparent gains hold up under careful evaluation.
Selected Work
Evidence, not just interests
A few current projects that show how I work: specify a question, build a system or study, and evaluate the result carefully.
Model adaptation
Bio-Informed LoRA for signal peptide prediction
Co-first author on a parameter-efficient adaptation method for ESM-2 that combines biological priors with multi-seed evaluation.
Reproducible evaluation
Hinglish named-entity recognition benchmark
Built a controlled comparison of fine-tuned multilingual models and zero-shot baselines on code-mixed language data.
Updates
News
- Jun 2026 Completed a working paper with Benjamin Rosche and Hanan Salam specifying an online experiment on strategic interpretation, normative interpretation, and AI-supported collective action in shared supply chains. The experiment remains under development. Project overview →
- 2026 Bio-Informed LoRA, our method for incorporating biological priors into ESM-2 adaptation, is under review at an EMNLP 2026 workshop. I am a co-first author. Publication details →
- Jan 2026 Joined NYU Abu Dhabi's eBRAIN Lab to work on parameter-efficient adaptation of ESM-2 protein models.
About
Research approach
I am completing a B.S. in Computer Science at New York University Abu Dhabi, with study-away terms at NYU Paris and NYU New York. I work across the full empirical cycle: formulating a question, preparing data, developing a model or system, evaluating it, and communicating the result.
With Benjamin Rosche and Hanan Salam, I am developing an online experiment on how AI-mediated interpretation and coordination affect collective action. In the eBRAIN Lab, I work on parameter-efficient adaptation of ESM-2, with an emphasis on benchmark design, sensitivity analysis, and replication across runs.
Research Interests
Methods and applications
Data and Statistical Analysis
Transforming complex and unstructured data into measurements, representations, and evidence that can answer substantive questions.
AI Experimentation and Evaluation
Designing controlled comparisons and using statistical inference to evaluate model behavior, efficiency, and downstream outcomes.
Applied Systems and Public Impact
Building reproducible computational tools and connecting technical results to decisions, human needs, and socially consequential problems.