KABIR MURJANI

Research

My work primarily focuses on reinforcement learning, game theory, and parameter efficiency these days. I am particularly interested in modeling adversarial dynamics, sequential decision processes, and building exact solvers for combinatorial spaces. Below is a selection of my recent publications and manuscripts currently under review. Whether you're building in adjacent spaces, or exploring topologies, feel free to reach out to me via email.

Peer-Reviewed Publications

Kabir Murjani, M. Bhavsar, M. Patel, J. Talukdar. "AlphaRoute: LLMs as Semantic Optimizers for Multi-Objective Routing." IEEE International Conference on LLM-Aided Design (LAD), Stanford University, 2026. (Oral)[arXiv][Code]

  • LLM-guided global router for VLSI physical design on the ISPD 2025 benchmarks: a language model tunes the PathFinder penalties while a knowledge graph keeps every suggestion inside the feasible design space.

Kabir Murjani. "Zero-Copy Semantic Contagion: An In-Memory Streaming Architecture for Evolving Attention Graphs." ACM SIGMOD Workshop on Data Management for the Modern Financial Systems (FinDS), 2026. (Oral)[arXiv][Code]

  • A zero-copy Rust engine for continuous-time contagion modeling over evolving attention graphs, with the streaming vector-similarity pipeline and benchmarking harness released alongside.

Kabir Murjani, Parth Vyas. "LURE: Bayesian Signaling Game in Multi-Step Agent Interactions." Conference on the Mathematics of Artificial Intelligence (MathAI), Sochi, 2026. (Oral)[OpenReview][Code]

  • Models how a long-running agent’s resistance to a bad deal erodes on its own, treating vulnerability as an endogenous utility deficit, and characterises the refusal policy an adversary learns against it.

Manuscripts Under Review

Kabir Murjani, Abhay Sobhanan. "Drive, Pack, Fly: The Travelling Thief Problem with Drone." Under review, European Journal of Operational Research (EJOR).[arXiv][Code][Data][Weights]

  • A capacitated truck whose velocity decays affinely with load, paired with a single-package drone launching and rejoining at route nodes. Exact MILP, metaheuristics, attention policies, and a learner-initialised hybrid.
  • Released with 219 benchmark instances, 60 trained policies, and the full result tables on Hugging Face.

Kabir Murjani, Abhay Sobhanan. "Neural Combinatorial Search Needs Rollouts." Under review, 18th International OPT Workshop on Optimization for Machine Learning, NeurIPS 2026.

  • On why learned construction policies underperform without search at inference time, and what the rollout budget actually buys.

Kabir Murjani, Nisarg Patel. "The Combinatorics of Trust over Financial Knowledge Graphs." Under review, 11th Workshop on Financial Technology and NLP (FinNLP), EMNLP 2026.[Code]

  • Budgeted evidence selection over temporal knowledge graphs with closed-form coherence scoring and a learned graph-transformer policy.

Kabir Murjani, Parth Vyas, Akshita Abrol, Rajesh Gupta, Z. Wang. "The Variance of a Judge: Zero-Variance Rewards by Deterministic Regression." Under review, ACL Rolling Review (ARR).

  • A judge-free alignment framework replacing a stochastic LLM judge with deterministic embedding regression, removing reward variance from the optimisation loop.
© 2026 Kabir Murjani.