University of California, Riverside
M.S. in Computer Science
Sep. 2024 - Mar. 2026
Riverside, CA
Coursework: DBMS (A+), Machine Learning, Artificial Intelligence, High Performance Computing, GPU Architecture, Advanced Networks
Software Engineer - Backend & AI Platforms
Mingyang Li
I build backend systems for document retrieval, traceable AI workflows, and reliable order processing.
M.S. Computer Science graduate focused on backend systems and AI platforms. My work spans reactive Spring services, hybrid document retrieval, citation validation, human-approved agent actions, and concurrent order flows with explicit cache-consistency trade-offs.
Fremont, CAActively seeking full-time Software Engineer roles in Backend, AI Platform, and Enterprise RAG.
Primary domain
Enterprise RAG
Backend proof
Orders + inventory
Retrieval layer
pgvector + RRF
Agent controls
Human approvals
Technical Skills
Languages
Java 17PythonSQLTypeScript/JavaScriptC/C++Go
Backend
Spring BootWebFluxProject ReactorR2DBCREST/SSE APIsRedisPostgreSQL/pgvectorMySQLKafkaMinIO
AI / Retrieval
OpenAI APIsParent-child RAGHybrid vector/full-text retrievalRRF fusionHeuristic rerankingCitation validationPi SDK
Tools / Infrastructure
ReactDockerLinuxGitGitHub ActionsAWS fundamentalsCI/CDDistributed-system debuggingJVM debugging
Education
Shanxi University
B.E. in Software Engineering
Sep. 2020 - Jun. 2024
Taiyuan, China
Expertise
Backend infrastructure for retrieval-heavy AI products.
My strongest work sits at the boundary between AI retrieval quality and classical backend reliability, including transactional order systems.
Ingestion to answer pipeline
Document RAG & Agent Backend
Connecting MinIO ingestion, parent-child document chunks, and OpenAI embedding and answer APIs through reactive REST/SSE services.
Vector + full-text + evidence
Hybrid Retrieval & Citation Context
Combining pgvector and full-text retrieval with RRF and reranking, preserving complete evidence spans within context budgets, and validating citations.
REST/SSE + checkout flows
Reactive & Transactional Backend APIs
Building Spring services for AI queries and checkout, with idempotency, order-state transitions, duplicate-message handling, and payment/cancellation race handling.
Cache isolation + human approvals
Scoped State & Agent Recovery
Scoping Redis context caches by tenant, actor, and document revision, and keeping durable agent state in Java with human approvals, stale-state checks, and conservative recovery.
Research Experience
Evaluation work for AI systems that need measurable robustness.
Graduate Research Assistant
RIPLE Research Group, UC Riverside
Advisor: Prof. Qian Zhang
Sep. 2025 - Present
Riverside, CA
- Built Python experiment runners and model adapters for GRAphRef, a constraint-guided fuzz-testing framework for 3D mesh AI models.
- Standardized evaluation across 8 mesh-processing systems including MeshCNN and HodgeNet, producing repeatable logs and metric reports.
- Ran structural mutation experiments for Valid Input Rate (VIR) and Semantic Preservation Score (SPS), then generated benchmark artifacts and LLM-output verification scripts using Benford/Zipf-style distributional checks.
Selected Projects
Systems projects shaped around retrieval, concurrency, and agent workflows.
PROJECT_01Backend
Jan. 2026 - Present
NexusAgent - Document RAG & Agent Backend
A document RAG and agent backend that makes retrieval, evidence selection, and agent actions inspectable, with a React/TypeScript learning UI and human-approved document operations.
JavaWebFluxPostgreSQL/pgvectorRedisMinIOReact/TypeScriptOpenAI APIsPi SDK
- Built a WebFlux ingestion and retrieval pipeline with MinIO, PostgreSQL parent-child chunks, child-only embeddings, and hybrid pgvector/full-text search fused through RRF and heuristic reranking.
- Integrated OpenAI embedding and answer APIs with REST/SSE, citation validation, and LLM-based follow-up resolution over bounded page history; visualized actual pipeline stages in React/TypeScript.
- Scoped Redis context caches by tenant, actor, and document revision, with TTLs, experimental semantic reuse, evidence revalidation, and graceful fallback; reused query vectors after semantic cache misses.
- Fixed evidence loss within a 4,000-character context budget by reserving complete child spans before fairly expanding parent context, exposing included evidence, trimmed parents, and budget exclusions.
- Built a Pi SDK worker for document diagnostics and action proposals, keeping durable run state and tool execution in Java with human approvals, stale-state checks, and conservative recovery of uncertain writes.
- Added header-based tenant filtering, audit events, and ingestion-job tracking; validated 366 backend tests plus frontend and worker suites in GitHub Actions with isolated services and mocked model responses.
PROJECT_02Backend
May 2025 - Aug. 2025
High-Concurrency Order & Inventory Backend
An order, inventory, and checkout backend focused on retry-safe transactions and concurrent lifecycle handling, including duplicate messages, payment/cancellation races, and cache-consistency trade-offs.
JavaSpring BootMySQLRedisMessage QueuesRPCDistributed SchedulingDocker
- Built order, inventory, and checkout services with MySQL/Redis, designing composite indexes for item lookup, order status, inventory checks, and user order history.
- Modeled pending, paid, cancelled, and expired states with idempotency keys and duplicate checks; implemented asynchronous expiration and compensation with queue retries and scheduled workers.
- Handled duplicate messages and payment/cancellation races, using cache-aside reads, delayed invalidation, and dedicated thread pools while documenting consistency and downstream-isolation trade-offs.
PROJECT_03Backend
Oct. 2025 - Present
AI Agent Engineering Lab
A Dockerized sandbox for comparing LLM backends and replaying prompt and tool-use workflows, backed by structured logs, lightweight CI, and documented failure cases.
DockerLinuxPythonLocal/Remote LLM BackendsAgent FrameworksGitHub Actions
- Built a Dockerized environment for local and remote LLM backends, comparing context-window behavior, memory usage, request latency, and deployment trade-offs.
- Wrote Python prompt and tool-call replay scripts with structured logs; maintained lightweight CI, environment checks, setup documentation, and known failure cases across personal experiments.
Curriculum Vitae
Resume-ready details, embedded.
Current resume with education, research, backend projects, and AI platform work.
Contact
Backend and AI platform work is where I am aiming next.
Actively seeking full-time Software Engineer roles in Backend, AI Platform, and Enterprise RAG. I am especially interested in teams where retrieval infrastructure, backend reliability, high-concurrency business systems, and applied AI meet real product constraints.