Research
PILOT
Recovery simulation for distributed systems, accepted at NSDI '26.
Available for Work
Based in La Jolla, California
Reliable AI systems, recovery, and infrastructure
Reliable AI infrastructure and dependable systems.
I am Angting Cai, a software engineer and systems researcher working on LLM infrastructure, recovery, distributed systems, and verification.
Systems
Phoenix
OS-supported high-availability restart framework, accepted at SOSP '25.
Industry
Amazon
Customer-facing SDE internship project shipped through production-safe processes.
Selected Projects
Technical work that shows how I think and build.
These projects emphasize systems implementation, fault tolerance, and practical AI infrastructure rather than broad product surface area.
Edge Computer-Use Agent (ECUA)
Built a fully local computer-use agent on Ubuntu using a perception-plan-execute-reflect loop around Qwen3-VL-8B served with vLLM.
- Hybrid perception with screenshots and accessibility trees
- Deterministic grounding and constrained action schema
- FP8, prefix caching, and CUDA graphs reduced time-to-first-token by about 65%
Sharded Key/Value Service with Paxos Groups
Designed a fault-tolerant sharded key/value storage system in Go with a shard master and Paxos-backed replica groups.
- Configuration management with a shard master
- Replica-group coordination using Paxos
- RPC design for smooth server-to-server communication
Pager
Implemented a pager that manages virtual memory for application processes with a simulated MMU and fixed-size page table.
- Address space creation, faults, destruction, and mapping
- Swap-backed and file-backed paging support
- FSM design for multicore scenarios
Intel P6-style Out-of-Order Pipeline
Implemented a scalar out-of-order pipeline in SystemVerilog with Tomasulo scheduling and a reorder buffer for VeriSimpleV.
- In-order commit with out-of-order execution
- Instruction prefetching and load-store queue enhancements
- Docker-based benchmarking against a baseline pipeline
Experience
Research depth with production-minded execution.
Research Assistant
Pilot Execution, University of Virginia
Studied recovery failures where the recovery action itself can trigger irreversible or cascading failures in distributed systems. Implemented core components of PILOT, an execution model for dry-run recovery simulation, and applied it to systems including Hadoop YARN.
Software Development Engineer Intern
Amazon
Completed a customer-facing internship project from design choices through implementation, team communication, testing, and safe production deployment. Worked inside standardized CI/CD, A/B testing, and integration-test pipelines while using internal AI tooling to accelerate learning and execution.
Research Assistant
Phoenix Project, University of Michigan
Built Phoenix, an OS-supported recovery framework that accelerates high-availability restarts by preserving long-lived process state while discarding transient state. Worked on Linux syscalls, allocator metadata, memory mapping management, runtime APIs, and fault-injection infrastructure for Redis and LevelDB validation.
Instructional Aid
Introduction to Logic Design, UM-SJTU
Held weekly labs, discussions, and office hours for a core ECE course with more than 100 students, helping improve hands-on implementation and debugging ability in digital logic labs.
Education
Strong academic grounding across systems and computer engineering.
University of California San Diego
M.S. Computer Science
GPA: 3.97 / 4.0
University of Michigan
B.S. Computer Science and Engineering
GPA: 3.88 / 4.0, Dean's Honor List 2022
Shanghai Jiao Tong University
B.Eng., UM Joint Institute
GPA: 3.53 / 4.0, Undergraduate Excellent Scholarship 2021
Research Highlights
High-signal work on recovery and dependable systems.
PILOT
Co-authored work on dry-run execution for recovery actions in production distributed systems, making it possible to observe consequences before commitment.
Phoenix
Co-authored work on OS-supported recovery for high-availability restarts.
View proceedingsSkills and Systems Stack
Research interests, languages, and infrastructure I work with.
Research interests
Reliable AI/ML systems
LLM inference
Agent infrastructure
Distributed systems
Recovery and fault tolerance
Systems verification
Languages
Python
C/C++
Go
Java
JavaScript
Shell
Systems and AI infra
Linux
Docker
AWS
Redis
LevelDB
Hadoop YARN
Paxos
vLLM
RAG
CI/CD
Fault injection
Prefix caching
CUDA graphs
FP8 optimization
Contact
Open to conversations about research, engineering, and dependable AI systems.
The fastest way to reach me is by email. If you are hiring for software engineering, systems, infrastructure, or research-adjacent roles, I would be glad to talk.