Advisor: Bryan Donyanavard

• Building Yggdrasil, a 3D scene graph for robots, written in Rust and designed for both construction and querying. Its layers form a DAG rather than a stack, and it answers pattern-matching, nearest-neighbor, field-of-view and traversal queries in place, using per-layer and per-node spatial indexes it keeps current as the graph changes.
• Every measured query takes 2–127 µs, up to 121× faster than a published C++ baseline. Yggdrasil is integrated into three published robotics pipelines and removes up to 99% of the time they spend on their scene graph operations.
• Currently extending Yggdrasil to multi-agent settings where each agent contributes its own view of the world, and a global server resolves the logical and semantic conflicts between them.
• Fully funded by NSF award CNS-2348349.

With Bogdan Alexandru Stoica

• AEBench measures how effective AI agents are in reproducing the results of published systems artifacts, and proposes a systematic framework for evaluation and execution of AI agents on artifact evaluation.
• The goal is to provide Artifact Evaluation Committees with a tool to fully automate the evaluation process via use of AI agents.

Advisor: Abbas Heydarnoori

• Built LinkAnchor, an autonomous agent that recovers links between issues and the commits that fix them (FSE ‘26, first author).
• LIA: supervised fine-tuning of LLMs for automatic issue assignment.
• Bug triaging with instruction-tuned LLMs.

• University merit fellowship awarded by SDSU for excellence in research and academic achievement, in addition to my NSF-funded research assistantship.

• Ranked 118th in the National University Entrance Exam on Mathematics (Konkour) among more than 144,000 students nationwide. (Top 0.08%)

• Awarded for creating Piqniq, an emergency alert app designed to support users through panic attacks.

• Code o She’r is a Spaghetti Code Challenge that focuses on undefined behavior and exploits architectural behaviors of computers and compilers with the aim of hosting an environment for contestants to become more familiar with the field of computer systems.
  
  
• As the Founder and President of Code o She’r, I took on the role of leadership over a team of 70+ members and oversaw various flows from design, marketing, and executive duties to scientific and technical tasks.

• Students’ Scientific Chapter (SSC) is a scientific association consisting of all students of Computer Engineering Department of Sharif University of Technology. It holds scientific talks and events throughout the year. Its central council consists of nine members, selected by votes of all students.
  
  
• I was elected vice president of technology of the SSC for one academic year, after receiving the most votes in the general election and then the votes of the other central council members.

• The ICPC International Collegiate Programming Contest, known as the ICPC, is an annual multi-tiered competitive programming competition among the universities of the world.

• As Chief of Technical Staff, I led a technical team of 10 members and addressed a variety of technical needs, such as:
  
  

  • Build contest-customized OS images and distribute them on contestants’ hosts using PXE Network boot.
  • Manage and configure hosts used by contestants and monitor their connectivity.
  • Restrict external access and host to host access of each contestant’s host device.
  • Ensure network performance and fairness between different devices used by the contestants.

• AIC is the biggest AI contest in the country, in which teams develop AI strategies to compete against other teams.

• As the Chief of Infrastructure and System Team, I was tasked with leading a team of 4 to address the Infrastructure needs of the contest:
  
  

  • Develop an event-driven matching system.
  • Design and Develop an auto-scaler system for the matching infrastructure to ensure minimum latency for the matches.
  • Ensure fair resource allocation between rival binaries.
  • Create an isolated sand-box for each contestant binary and eliminate any side-channel attack that may affect the fairness of the results.
  • Maintain data storage for everything from game logs to contestant binaries and monitoring data.

Sotoon is a B2B cloud provider that focuses on reliable and scalable computing solutions. I worked in the K8s-as-a-Service team, designing and maintaining Sotoon’s Kubernetes engine and several integrations to Sotoon’s services, such as CSI, CNI, and Auto-scaler.
Key Achievements:

• Automated management of 90+ Kubernetes clusters, optimizing operational efficiency and reducing tickets by 30%.
• Achieved a 40% reduction in initial cluster costs by eliminating duplicated system components and introducing multi-tenancy.
• Redesigned cluster network topology to enhance security and abstract control plane management from customers.
• Planned and led the migration of over 100 customers’ k8s-clusters to the new infrastructure with zero downtime.
• Implemented an A/B testing framework for cluster updates, preventing 5 otherwise catastrophic failures.
• Investigated and debugged a critical data race issue in the CSI plug-in that could have resulted in the deletion of over 70TB of client storage data, ensuring data integrity and service reliability.
• Automated the CI/CD pipeline for OS image creation, reducing new k8s version release time from 1 week to 1 day.

Digikala is the largest e-commerce company in the Middle East, serving over 40 million users. As part of the Big Data & AI team, I was responsible for building and maintaining the infrastructure required to efficiently store, process, and manage the vast amounts of data generated by user activity.
Key Achievements:

• Developed a Parallel Data Query Engine on top of Apache Spark, revising data retrieval models and queries to reduce resource usage by 23% and computation time by 43%.
• Reduced API latency by identifying blocking operations in an asynchronous workflow, optimizing the critical path to triple requests per second (RPS).
• Designed and maintained a data platform using K8s for orchestration, Ceph for storage, Airflow for scheduling, Spark for batch/stream processing, and Kafka with Debezium for database synchronization.
• Implemented GitOps-based infrastructure management with ArgoCD, automating API deployments via Kustomize and Helm, reducing release cycles from 6 months to 6 hours.
• Deployed and maintained a staging environment and built a framework for the Data Science team, reducing their development time significantly.

Tapsell is a leading AdTech company in the Middle East. As part of the DevOps team, I provided cloud infrastructure and platform services that enabled other technical teams to build, deploy, and scale their applications efficiently.
Key Achievements:

• Optimized Cassandra’s disk usage for different workloads, resulting in a 90% reduction in disk usage for some of our clusters.
• Designed and implemented a fully automated backup, validation, and restore system for our cloud-native infrastructure, replacing the legacy solution while preserving backward compatibility to enable seamless, incremental migration of 30TB+ of data with zero downtime.
• Built and maintained infrastructure by deploying a staging cluster on K8s, developing CI/CD pipelines with GitLab CI, automating provisioning with Terraform on VMWare, and administering Linux-based VMs to resolve network bottlenecks and improve performance.