Go/Python Event-Driven Platform & LLM Engineer

Lihan Zhan

***********@*****.*** • 917-***-**** • Linkedin.cg • github.com/Cchariot

Work Authorization: US Citizen

HIGHLIGHT

Software engineer with handson experience in Go/Python backend development, event-driven data platforms, and Kubernetes deployments, plus a strong background in machine learning, generative AI, and edge inference. Skilled in TypeScript, React, PostgreSQL, NATS JetStream, Docker, Kubernetes, Helm, AWS, and CI/CD.

EDUCATION

Carnegie Mellon University, Pittsburgh, PA 01/24-12/25

Master of Software Engineering Core GPA: 3.95

The City College of New York, New York, NY 08/16-05/21

Bachelor of Science, Computer Science Core GPA: 3.80

TECHNICAL SKILLS

•Programming Languages: Go, Python, TypeScript, C++, Java, R, SQL

•Tools and platforms: AWS, Docker, Kubernetes, Helm, NATS JetStream, Apache NiFi, PostgreSQL, Istio, PyTorch, OpenAI API, Git, Bitbucket, JIRA, Jenkins, Robot Framework, Linux, Anaconda

•Web Technologies: React, TypeScript, JavaScript, Node.js, HTML

WORKING EXPERIENCE

Software Engineer II at Software Engineering Institute of CMU, Pittsburgh PA 05/2022-Current

•Built a self-service data platform on Apache NiFi that lets non-technical operators ingest data sources, chain containerized transformations, and send output to external systems, with NATS JetStream to ensure delivery.

•Built the platform's operator-facing web frontend in React and TypeScript.

•Packaged and deployed platform services to Kubernetes with Helm charts, behind an Istio service mesh with Keycloak single sign-on and Kyverno admission policies.

•Built an agentic LLM pipeline (tool calling, retrieval) that generates executable MLTE test suites with grounding checks.

•Built an LLM-based pipeline to translate legacy Ada code to C++, increasing translation throughput by 4x.

•Deploy edge-optimized LLM inference pipeline using Hugging Face and ONNX Runtime, reducing response latency by 40% on Jetson Nano.

•Designed and implemented a real-time facial recognition pipeline using PyTorch, deployed on AWS with S3 for scalable image storage and a PostgreSQL database for metadata indexing.

•Implemented GPU-accelerated image processing (CUDA) in ROS pipelines on Jetson to automate drone missions.

•Created a container minimization tool, shrinking image output tar size by ~90%.

•Technology and tools used: Go, Python, TypeScript, React, Apache NiFi, NATS JetStream, Kubernetes, Helm, Docker, AWS, PyTorch, OpenAI API, CUDA, ROS.

Software Engineer at Lockheed Martin, Moorestown, NJ 07/21-05/22

•Improve Solid State Radar performance and meet client special needs using C++ and Python.

•Set up CI/CD using Jenkins & Robot Framework, reduced average test run time by 30%.

•Develop software with Agile Scrum methodology, use JIRA and Bitbucket for team collaboration.

•Identify radar performance change and error using GDB

•Technology and tools used: C++, Jenkins, Robot Framework, JIRA, Bitbucket, GDB.

SELECTED PROJECT

Small Language Models on Edge Devices (Individual) 05/25-11/25

•Built a stress-test benchmark on Jetson Xavier NX (8 GB) for 6 small language models (82M to 360M params), sweeping prompt length from 128 to 512 tokens until timeout, OOM, or crash.

•Measured latency per token, energy per token (tegrastats power traces), peak GPU memory, model size on disk, and failure events at each prompt length.

•Found that models up to 125M params stayed interactive at 512 tokens, while 350M+ models became unstable past ~320 tokens; SmolLM-135M used ~4x the latency, energy, and memory of DistilGPT2 at 512 tokens.

•Turned the results into a deployment playbook (treat prompt length as a budget, use models up to 125M for interactive tasks, shorten prompts or quantize before scaling to 360M) and a reusable stress-test template for new devices and models.

Contact this candidate