Senior Platform Engineer · Boston, MA
Johnson Wu
I build the platform your team ships on.
Platform engineer who ships hard things fast — LLM pipelines, real-time data infrastructure, and developer platforms that make entire engineering orgs move faster.
Cost reduction
LLM localization cut $100K/release down to $1K
Faster releases
Artifact portal cut 2-hour cycles to under 2 minutes
What I Build
Full-stack platform tools — GitLab CI / ArgoCD GitOps pipelines, Grafana + Loki observability across 10+ production services and 50+ build machines.
Platform Engineering
Internal developer platforms, GitOps pipelines, and observability stacks that scale engineering teams.
LLM & AI Infrastructure
Production LLM pipelines with FastAPI + Celery, RAG systems, and agentic workflows with LangChain/LangGraph.
DevOps & Cloud
CI/CD automation, containerized infrastructure, and scalable cloud deployments on AWS and Kubernetes.
Full-Stack Development
End-to-end web apps, REST APIs, and fault-tolerant ETL pipelines — React frontends to scalable backend systems.
Experience
- Spearheaded an LLM-powered localization service (FastAPI, Celery, Redis, SQL) — cutting per-release costs from $100K to $1K, a 99% cost reduction; implemented token-level cost tracking.
- Drove end-to-end development of a real-time artifact publication portal — cutting release time from 2 hours to under 2 minutes and eliminating hours of manual verification.
- Built fault-tolerant ETL pipeline with parallel partition workers, transactional rollback, and idempotent retries — reliably ingesting 100M+ rows from legacy databases and CI/CD servers.
- Implemented full-stack source-control automation tool (Flask, Redis, SQL, React) — automated Perforce branching and stakeholder notifications, reducing hours of manual workflows into seconds.
- Building full observability stack on Kubernetes (Grafana, Loki, OTel Collector) — centralized logging across 10+ production services and 50+ build machines.
- Architected 50+ secure API endpoints; established GitLab CI + Ansible CI/CD framework, progressively migrating to GitOps with ArgoCD.
- Led development of end-to-end test automation frameworks for AR/VR and Android (C#, Python, Jenkins, Jira) — cut release test time by 80% with 50+ robust test cases.
- Rebuilt Jenkins multibranch pipelines with Dockerized agents on Kubernetes — cutting build spin-up time by 99%.
- Designed and implemented a hybrid cloud ML pipeline for banking data residency using AWS SageMaker, CodePipeline, and EventBridge.
- Created CloudFormation templates enabling full infrastructure deployment in under 10 minutes.
- Applied GitOps with ArgoCD on Kubernetes clusters for automated sync and monitoring.
- Built Transformer-based Taiwanese-Chinese translation model achieving BLEU score of ~90 on 10,000+ sentences.
- Deployed translation models as a Streamlit web app with real-time parameter adjustment.
- Led 4-person team fine-tuning GPT-2 for commercial dialogue rephrasing.
Education
Certifications
Projects
Open Source
LangChain Anytype Document Loader
Open-source LangChain document loader for Anytype — sync/async APIs, filtering, batch processing. Featured in official Anytype developer docs. Also shipped a bug fix merged into flask-pydantic v0.14.0.
Open Source
Serverless AI Assistant
LangGraph-powered agentic AI assistant on AWS Lambda + Supabase, integrated with LINE messaging API. Full IaC with Terraform + GitHub Actions CI/CD.
Web Development
Owl Go — Social Networking
Full-stack social networking site with OAuth 2.0, account linking, post/comment system, follows, and image uploads.
Machine Learning
Taiwanese-Chinese Machine Translation
Transformer model (Hokkien → Traditional Chinese) with Beam Search + Length Penalty. BLEU ~90 on 10K+ sentences.
Machine Learning
CLIPGPT-ImageCaptioner
GPT-2 + CLIP image captioning using ClipCap architecture — CLIP encoding as prefix to fine-tuned GPT-2 for natural captions.
Desktop Application
Chinese Language Sample Analysis
PyQt5 desktop app for speech therapists — word segmentation, POS tagging, speech-to-text, and MLU analysis with CKIP Transformers + Azure Speech SDK.