ROLE // Agent Engineer▍

AI Engineer building Agent and LLM products.

Production CV/ML background — ex-Tencent YouTu, PhD in ML.

View work Resume GitHub LinkedIn

2,200+

GitHub stars

1,400+

product users

CVPR · IEEE

published research

ICCV LVIS

challenge champion

Profile · 00 Available

Tianliang Zhang · Jay

AI Engineer · Agent & LLM

LOCATION Canberra, Australia GEO 35.28°S 149.13°E PhD Signal & Info Processing

01What I build

Selected AI tools and LLM products, shipped from idea to real users.

★ Agent tooling

Skills Manager

Open-source desktop app for managing AI-agent skills across 15+ coding tools.

Built a cross-tool skill manager for Cursor, Claude Code, Codex, Copilot and other AI coding environments.
Reached 2,200+ GitHub stars and 190+ forks as an open-source developer tool.
Shipped a Rust/Tauri core with a React interface for local-first skill organization and sync.

RustTauriReactTypeScript

Source Demo

open-source

2,200+

GitHub stars · 190+ forks

15+ coding agents · Cursor · Claude Code · Codex · Copilot

LLM product 02

EasyIELTS

AI IELTS speaking and writing coach with scoring, feedback and 24/7 practice.

Built an AI coach for IELTS speaking and writing with rubric-based feedback.
Delivered fluency, vocabulary, grammar and pronunciation scoring with sentence-level suggestions.
Served 1,400+ users with 1,090+ questions and 1,300+ audio examples.

·1,400+ users·Live

Next.jsLLMSpeechSupabase

Live

Generative AI SaaS 03

Sticker It

Generative-AI SaaS that turns photos into custom stickers.

Launched a paid AI sticker generator with image upload, generation, checkout and delivery flows.
Integrated OpenAI, Stripe, S3, email, Sentry and Docker for a production SaaS stack.
Designed the end-to-end user journey from photo input to downloadable sticker output.

·Paid SaaS·Stripe

Next.jsOpenAIStripeAWS S3

Live

LLM engineering 04

ACT Learner

LLM-powered road-rules coach with cached generation and multi-provider fallback.

Built a bilingual ACT road-rules learning coach with explanations, scoring and mistake review.
Reduced live-LLM dependency with prompt-hash seeding, cached generations and precomputed content.
Implemented OpenAI/Gemini fallback for resilient production-grade answer generation.

·Live·Multi-provider

Next.jsOpenAI/GeminiPrompt Eng

Live

02AI Engineering Stack

Turning agent and LLM ideas into reliable products takes more than model calls.

01 Agents

Agent tooling & workflows

Skill libraries, adapters, sync and project-scoped workflows across coding agents.

Skills Manager · 16+ tools · 2.2K stars

AgentsToolingRust/Tauri

02 Structured output

LLM scoring & structured output

Turned subjective grading into evaluable, parseable and iterated strict-JSON contracts.

EasyIELTS · 1,400+ users

Structured outputEvalsEdTech

03 RAG

RAG & grounded answers

Chunked official source material, mapped questions to sources and gated unverified spans.

ACT Learner · official handbook grounding

RAGGroundingQuality gates

04 Caching

Cost, latency & fallback

Used caching, pre-generation and model fallback to make LLM features production-safe.

Prompt-hash cache · OpenAI/Gemini fallback

CachingReliabilityFallback

05 Next.js

Full-stack AI product delivery

Shipped product design, model integration, payments, deployment and growth end to end.

4 shipped AI products

Next.jsSupabaseStripe

06 ONNX

Production AI / ML systems

Delivered high-accuracy, low-latency inference systems across identity, inspection and edge AI.

Sub-second ONNX · 48 machines · edge RK3566

ONNXEdge AIDeployment

03Engineering foundation

Identity verification, robotics and inspection work framed as deployable, measurable production AI systems.

2025 — Now

LexisNexis Risk Solutions

Senior Software Engineer · Production AI Systems

Shipped document-preprocessing and forgery-detection systems for global identity verification; re-architected ONNX inference for sub-second latency and production-grade accuracy.

Production AIONNXSub-second

2023 — 2024

PICEA (3irobotix)

Senior Algorithm Engineer · Edge AI

Built an edge-AI perception stack for voice/gesture-controlled robots — pose, detection and LiDAR-fused ranging; lifted YOLO inference from 6 to 30 FPS on RK3566.

Edge AI6→30 FPSROS

2020 — 2023

Tencent · YouTu Lab

Senior Researcher

Moved industrial AI inspection from algorithms to production lines — high-recall defect detection across 48 machines processing 24K cells/day (~188GB).

Industrial AI48 machinesPyTorch

2021

ICCV 2021 · LVIS Challenge

Long-tail Instance Segmentation — Champion

Swin Transformer + CBNetV2 backbones with a novel Seesaw Loss and augmentation; Rank 1 and Best Innovation Award.

ChampionBest InnovationSwin

04Research record

Earlier computer-vision research — robust detection, large-scale recognition and benchmark-winning systems. The rigor behind the engineering I do today.

CVPR 2020 · co-first author

Multiple Anchor Learning for Visual Object Detection

Object-detection architecture for dense visual recognition.

PDF

1st Place · ICCV 2021 · co-author

LVIS Challenge 1st-Place Solution: Distribution Balanced & Boundary Refinement for Large-Vocabulary Instance Segmentation

Award-winning large-vocabulary instance-segmentation system.

PDF

IEEE TITS 2020 · first author

Feature Calibration Network for Occluded Pedestrian Detection

Robust pedestrian perception under heavy occlusion.

PDF arXiv

View all publications · 9

Patents

2 US patents · 13 China patents

Google Scholar

Awards

ICCV LVIS Champion
Tencent Outstanding Employee
CSIG Tech Breakthrough
National Scholarship

05Writing

Engineering notes from real projects — production LLM systems, coding agents and computer vision.

11 June 2026 LLMEvals

Grade the grader: an eval harness for an LLM IELTS examiner

My last post claimed a rubric-as-contract makes LLM grading reproducible. This one tests that claim on real production data — and finds the grader is a distribution, the fallback model is +0.34 band more lenient, and top scores drifted almost a full band in three weeks.

9 June 2026 LLMAI Engineering

Seed-first AI: serving LLM explanations at near-zero cost

How ACT Learner serves grounded AI explanations to real users while calling the live model on only a fraction of requests — and why the prompt is the cache key.

7 June 2026 AI AgentsDeveloper Tools

One library, every agent: syncing AI-agent skills across 15+ tools

Coding agents each store skills in their own folder, their own way. Skills Manager normalises all of them behind one central library — the architecture, and why it's a Rust filesystem program wearing a GUI.

All writing

06Let's talk

Interested in AI agent tooling, LLM products, or production ML systems?

● Open to AI Engineer / Agent Engineer roles.

tianliangjay@gmail.com Resume LinkedIn GitHub X