Kia Rahmani

Kia Rahmani

Applied Artificial Intelligence Scientist

Kia Rahmani

I am an applied scientist working at the intersection of programming languages, formal methods, and modern AI systems.
My recent work focuses on LLM-based program synthesis, agent orchestration, and evaluation infrastructure for real-world applications.

I’m a founding engineer at Durable, where we work on reliable agent/program synthesis from product requirement documents, multi-API integration, and automated QA/repair pipelines for LLM agents.

Before that, I was a post-doctoral researcher at the University of Texas at Austin with Isil Dillig and Joydeep Biswas, and I received my Ph.D. in Computer Science from Purdue University with Suresh Jagannathan and Ben Delaware. I also spent time at Microsoft Research working with Sumit Gulwani on LLM-guided program synthesis in 2020.

I like problems that combine abstract reasoning (logic, semantics, verification) with messy real systems (databases, agents, APIs, robots).

What I work on

  • LLM agents & program synthesis – building systems that translate natural language into reliable, debuggable code and workflows.
  • Integration & infra – connecting LLM agents to external APIs (Slack, Google, Salesforce, etc.), with testing, logging, and automated repair.
  • RAG, QA, and evaluation – pipelines that simulate real users, surface failures, and automatically suggest fixes.
  • Neurosymbolic & formal methods – combining temporal logic, model checking, and program analysis with learning-based systems.
  • Distributed data & consistency – reasoning about weakly consistent databases and refactoring schemas to eliminate subtle concurrency bugs.

Selected publications & patents

  • Dynamic Model Predictive Shielding for Provably Safe Reinforcement Learning
    A. Banerjee, K. Rahmani, J. Biswas, I. Dillig
    NeurIPS 2024
    pdf

  • Programming-by-Demonstration for Long-Horizon Robot Tasks
    N. Patton, K. Rahmani, M. Missula, J. Biswas, I. Dillig
    POPL 2024
    pdf

  • Programmatic Imitation Learning from Unlabeled and Noisy Demonstrations
    J. Xin, L. Zheng, K. Rahmani, J. Wei, J. Holtz, I. Dillig, J. Biswas
    IEEE RA-L
    pdf · Project · Video

  • Multi-modal Program Inference: a Marriage of Large Language Models and Component-based Synthesis
    K. Rahmani, M. Raza, S. Gulwani, V. Le, D. Morris, A. Radhakrishna, G. Soares, A. Tiwari
    OOPSLA 2021
    pdf

  • Repairing Serializability Bugs in Distributed Database Programs via Automated Schema Refactoring
    K. Rahmani, K. Nagar, B. Delaware, S. Jagannathan
    PLDI 2021
    pdf

  • CLOTHO: Directed Test Generation for Weakly Consistent Database Systems
    K. Rahmani, K. Nagar, B. Delaware, S. Jagannathan
    OOPSLA 2019
    pdf

  • Multi-modal Program Inference (US20230176829A1)
    K. Rahmani, M. Raza, S. Gulwani, V. Le, D. Morris, A. Radhakrishna, G. Soares, A. Tiwari
    US Patent Application
    Patent

  • Symbolic Analysis of Weak Concurrency Semantics in Modern Database Programs
    K. Rahmani
    PhD Thesis, Purdue University, 2022
    Thesis

Academic service

  • Program Committee
    • OOPSLA 2026 — ACM Conference on Object-Oriented Programming, Languages, Systems, and Applications
    • ICLR 2025 — International Conference on Learning Representations
    • ICML 2025 — International Conference on Machine Learning
    • NeurIPS 2024 — Conference on Neural Information Processing Systems
    • DoE SBIR/STTR Program
    • IEEE RA-L — Robotics and Automation Letters
    • TAHRI 2023 — International Symposium on Technological Advances in Human-Robot Interaction
    • NeurIPS 2021 workshop on Advances in Programming Languages and Neurosymbolic Systems (AIPLANS)
    • IROS — IEEE/RSJ International Conference on Intelligent Robots and Systems

Contact

The easiest way to reach me is by email:

I’m open to roles and collaborations around LLM systems, AI infrastructure, agent orchestration, program synthesis, and verification-heavy systems work.