About
I am a final year Electrical Engineering student at Nirma University. I architect big neural nets and exact solvers that run fast.
I primarily work on reinforcement learning on combinatorial spaces and language models. I'm currently working on Deep RL solvers for NP-hard combinatorial problems at Indian Institute of Management Bangalore (IIMB) advised by Prof. Abhay Sobhanan, and previously engineered and optimized sequential architectures with SIT-NVIDIA AI Centre (SNAIC).
Mostly I like to work on deep learning, numerics, and systems. Among other things, I play chess and tennis, and compose algorithmic music. If you're building in adjacent spaces, feel free to reach out.
Interests: Deep Reinforcement Learning, Neural Combinatorial Optimization, Language Models, Decision Sciences.
Positions & Education
Recent Updates
2026
- Released the TTP-D benchmark suite: 219 instances, 60 trained policies, and the full solver result tables, on Hugging Face.
- Kabir Murjani, Abhay Sobhanan. Drive, Pack, Fly: The Travelling Thief Problem with Drone. Under review at Computers & Operations Research.
- Kabir Murjani. AlphaRoute: LLMs as Semantic Optimizers for Multi-Objective Routing. Accepted (oral) at IEEE LAD 2026, Stanford University.
- Volunteering at ACL 2026 in San Diego as Session Chair for the Main and Findings track.
- Selected as recipient of the ACM SIGMOD/PODS 2026 India Fellowship.
- Kabir Murjani. Zero-Copy Semantic Contagion: An In-Memory Streaming Architecture for Evolving Attention Graphs. Accepted (oral) at ACM SIGMOD FinDS, 2026.
- Started a research internship on deep reinforcement learning for neural combinatorial optimization at the Indian Institute of Management Bangalore, advised by Prof. Abhay Sobhanan.
- Elected General Secretary of the IEEE Computer Society chapter at Nirma University.
- Kabir Murjani, Parth Vyas. LURE: Bayesian Signaling Game in Multi-Step Agent Interactions. Accepted (oral) at MathAI 2026, Sochi, Russia.
- Released the ChessNano weights — a 51.9M-parameter SAN transformer in a 58.4 MB INT8 artifact — on Hugging Face.
- The Variance of a Judge: Zero-Variance Rewards by Deterministic Regression submitted to ACL Rolling Review.
2025
- Collaborating with Dr. Timothy Liu at the NVIDIA AI Center at Singapore Institute of Technology on quantized edge inference and adaptive routing architectures.
- Machine Learning Researcher at Xtin Capital, building a high-frequency news ingestion engine and an adversarial backtesting environment.
- Lead Aniate Labs to explore open source applied AI tooling at the intersection of language models and combinatorial solvers.
- Presented early results on DEED manifold compression at an internal seminar at Nirma University's EE department.
- Released the CRAFT-5 dataset: 2,384 RLAIF examples for preference tuning, on Hugging Face.
2024
- Joined Team Dyaus at Nirma University, building a CanSat model for the Türksat Satellite Competition 2024.
- Finished 3rd place at the Indian Institute of Technology Kanpur Chess Masters Premier League 2.0 among 60+ universities.