Adarsh Prabhakaran

Postdoctoral Researcher at RHUL. LLMs, agent-based simulation, and computational social science.

LLM Behavioural Fine-tuning

Parameter-efficient fine-tuning of small language models on behavioural datasets to produce psychologically grounded personas. Ongoing.

LLM Reasoning Survey

Studying reasoning in LLMs and benchmarking against human participants. Ongoing.

Adversarial multi-agent pipeline where LLM agents playing competing counsel personas stress-test legal arguments via a simulated judge; identified weaknesses feed a structured refinement loop. Evaluated on ECHR case databases. In collaboration with Ondrej Bohdal.

ECHR Compliance ABM

Agent-based model of state compliance behaviour in international human rights law, developed in collaboration with UCL and KCL. Integrates qualitative interview data, expert coding, and quantitative case records to deliver predictive compliance metrics and policy-relevant insights.

Tobacco Control Network Models

Multi-agent simulation framework modelling tobacco use dynamics on social networks. Designed graph-based intervention algorithms outperforming standard methods; ran large-scale parameter sweeps on HPC (Eddie). Explored spatial and network-based policy scenarios for Scotland.