About
I’ve spent 6+ years building backend systems, from chatbot and NLU products to distributed services and data platforms. At AmazingTalker I led a 3–4 person engineering team; at Sourect I owned backend architecture for a B2B MVP.
Now I apply the same discipline to LLMs, RAG and coding agents, while completing a Master of IT at QUT. I’m looking for backend or applied AI roles.
Work rights
I’m in Brisbane on a student visa with work rights now. After my Master finishes in Nov 2026 I move to a Temporary Graduate visa (subclass 485) with full, unrestricted work rights, so no employer sponsorship is needed in the near term.
Case studies
-
2021 – 2023
Data warehouse & developer platform AmazingTalker
Marketing and product teams waited on engineers for every report. I built a BigQuery warehouse with self-service reporting, and an Nx platform that deploys only the services a change affects.
- 30% → 5% of sprint time spent on reporting
- 3 days to ship a new data API, down from about a week
How it was built
- Given
- Business data, advertising data and in-house tracking lived in separate systems. Marketing and product teams depended on engineering work to prepare reports.
- Warehouse
- BigQuery with Datastream CDC, Airflow and dbt, plus Metabase for self-service reporting. It unified business data with GA, Meta, Google Ads, TikTok and in-house tracking.
- Platform
- An Nx monorepo with a service generator, shared libraries and CI/CD that built and deployed only affected services to EKS. Product teams adopted it to decouple services from the main app.
-
2021 – 2023
Teacher ranking & video pipeline AmazingTalker
New sales took about 20 minutes to reach teacher rankings. I replaced full batch recomputation with event-driven, per-teacher updates, and launched the short-video pipeline.
- 20 → 2 min for new sales to reach rankings
- −60% ranking API latency, 500 → 200 ms
- 1,000+ video transcoding jobs per day
How it was built
- Ranking
- Elasticsearch scoring and Redis sorted sets per locale. Event-driven updates recalculated individual teachers instead of the full batch.
- Video
- S3, SQS fan-out, ffmpeg on Lambda and HLS via CloudFront. Playback tracking fed teacher ranking and video recommendations.
-
2023 – 2025
RAG-based knowledge system Freelance
Staff needed to query private documents in plain language without content leaking between users. I built a RAG pipeline that answers only from permission-scoped context.
How it was built
- Method
- Automated ingestion, chunking and embedding into a vector store, with an LLM answering only from retrieved, permission-scoped context.
- Decision
- Per-user data isolation was part of the design from the start, so security held up as the document set and user base grew.
Experience
-
2026 – Now
Full Stack Engineer QUT Industry Project
Primary developer in a six-person team building an on-premise government AI proof of concept.
-
2024 – 2025
Contract Software Engineer (Part-time) Sourect
Owned the backend of a B2B garment-manufacturing MVP, including Stripe payments and an artwork validation and vectorisation pipeline.
-
2023 – 2025
Full Stack Engineer Freelance
Permission-scoped RAG, layout-preserving translation for 100+ page financial reports, and an iGaming CRM used by dozens of operations staff.
-
2021 – 2023
Senior Backend Engineer AmazingTalker
Led a 3–4 person team across data, ranking and service-platform work. Built an onboarding project used by ~10 engineers, typically shipping to production within 3 days.
-
2019 – 2021
Backend Engineer WEI-PO
Removed Redis KEYS scans and queued settlement, cutting API time from 400 to 50 ms. Built an idempotent wallet load-tested to 2,000+ QPS.
2 earlier roles, 2017 – 2019
-
2018 – 2019
Backend Engineer Pentium Network Technology
Integrated CDN and FQDN resources across AWS, Alibaba Cloud, Tencent Cloud and GCP; built chatbot and workflow automation with Kafka.
-
2017 – 2018
Backend Engineer Lingtelli
Built the company’s chatbot product for Facebook and LINE and its natural-language-understanding engine.
Projects
-
RAGcat
Upload PDFs, review the extracted passages and embed a support chatbot on your own site. Every answer cites the document and page it came from.
-
Codex Cage
Takes a GitHub or Linear issue to a pull request inside a disposable sandbox. Tests, a secret scan and an independent review must pass first.
-
Strobe Assistant
A conversational assistant for your own posts and photos. Chat stays separate from asynchronous media work, and mutations need explicit permission.
-
Knowledge Compiler
Turns scattered notes into a knowledge base. Every proposed change carries its source evidence, and you decide what gets applied.
-
AI Meeting Management Platform
An on-premise government proof of concept from a QUT industry project, turning recordings into draft minutes and action items.
More
Built for the Department of Science and Technology (DOST), Philippines, with me as the primary developer in a Brisbane-based team of six. Whisper transcribes recordings and a local Llama 3 model drafts minutes and action items for staff to review on the client’s infrastructure. A transactional outbox and Redis Streams provide durable job dispatch across Node.js and Python workers; SSE reports progress without blocking the API.
Education
-
2025 – 2026
Master of Information Technology QUT, Brisbane
Expected Nov 2026.
-
2011 – 2015
Bachelor of Computer Science NTUST, Taiwan
Skills
- Languages
- Go, Python, TypeScript / Node.js, Ruby, C#, NestJS, Express, FastAPI, React, REST / GraphQL
- Data
- PostgreSQL, MySQL, Redis, Elasticsearch, BigQuery, dbt, Airflow, CDC, RabbitMQ, Kafka
- Cloud
- AWS, GCP, Docker, Kubernetes (EKS), GitHub Actions, New Relic, Nx
- AI
- RAG, vector databases, Llama 3, Whisper, Amazon Bedrock, agentic workflows, MCP / ACP, Claude Code
Contact
Open to senior backend and applied AI roles, in Brisbane or remote. The fastest way to reach me is email.
- [email protected]
- linkedin.com/in/jhowliu
- GitHub
- github.com/jhowliu
- Résumé
- PDF, one page