Personal Portfolio & Resume

Liu Wei

AI & Full-Stack Developer  |  Computer Science Student

Solid foundation in AI theory with extensive project experience, specializing in deep learning, computer vision, and large language model applications. Complete experience from model selection, optimization, deployment to full-stack AI application development. Passionate about solving technical challenges in high-performance AI systems, eager to create value in real business scenarios.

sunflowerslw0607@gmail.com / +60 18-782-6582 / LinkedIn / GitHub
01

Education

Asia Pacific University (APU) & De Montfort University (DMU), Malaysia

2024 – 2027 (Expected)
Bachelor of Science (Hons) in Computer Science (Dual Degree)
  • CGPA: 3.88 / 4.00 (First Class Honours)  |  Entrance Scholarship Recipient  |  VC List Awardee
02

Hackathon & Competition

Deriv AI Hackathon 2026 — Finalist (Top 4–7)

Feb 2026
Team: TradeIQ  |  Cyberjaya, Malaysia  |  Core Developer & AI Pipeline Architect
  • Built "TradeIQ", an AI-powered intelligent trading platform for retail traders — awarded RM 2,000 prize as Finalist
  • Designed a 5-agent AI pipeline (Market Monitor → Event Analyst → Portfolio Advisor → Behavioral Sentinel → Content Creator) using DeepSeek V3 with streaming function calling
  • Developed flagship "Trading Twin" feature: dual equity curve simulation comparing impulsive vs. disciplined trading, with AI-generated behavioral narratives
  • Tech Stack: Next.js 16, React 19, Django 5, Django Channels (WebSocket), Supabase PostgreSQL, Upstash Redis, Deriv WebSocket API, Bluesky AT Protocol
  • Delivered full-stack product in 24 hours: real-time market analysis, behavioral coaching with risk scoring, and social content engine with one-click publishing
03

Internship Experience

Kuayue Xingkong (Hangzhou) Technology Co., Ltd.

Sep 2025 – Feb 2026
Large Language Model Algorithm Engineer
  • LLM Application & Optimization: Prompt engineering, CoT logic; Fine-tuned GPT, Gemini, DeepSeek; Built RAG with LangChain/LlamaIndex; LoRA/QLoRA/PEFT; Optimized inference (TensorRT, vLLM, DeepSpeed)
  • Agent System Development: Multi-agent architecture (Planner/Executor/Tool-Using Agent); Agent memory systems; Function Calling/Tool Use integrating APIs, databases, code execution
  • Backend & API: Python (FastAPI/Flask) + Node.js (Express/NestJS); PostgreSQL/MongoDB/Redis; Async queues (Celery/asyncio); Streaming (SSE/WebSocket)
  • Frontend & DevOps: React/Next.js/Vue/TailwindCSS; Docker/K8s CI/CD; AWS/GCP/Vercel; Defined AI product architecture from scratch
04

Project Experience

SynCVE — Real-time Emotion Recognition & Anti-Spoofing System

2025 – Present
Core Developer & AI Algorithm Lead
  • Integrated DeepFace and FasNet anti-spoofing model, achieving 85–95% accuracy for 7 core emotions and efficient liveness detection
  • Configured CUDA 11.8 GPU acceleration for TensorFlow & PyTorch; reduced detection latency from 15s to 1–2s (87% improvement)
  • Diagnosed and resolved PyTorch CUDA initialization failure (WinError 127); created environment configuration documentation

SynPod — AI-Driven Intelligent Travel Voice Assistant Platform

2025 – Present
Core Developer & AI Algorithm Lead
  • Integrated OpenAI GPT-4 (Function Calling) and Deepgram Nova-2 (STT) / Aura-2 (TTS) APIs for real-time voice interaction and podcast generation
  • Designed production-grade Token Bucket rate controller and FSM with 5 states for complex generation workflows
  • Achieved <300ms latency voice interruption via persistent WebSocket, intelligent text chunking, and content prefetching

SynText — AI-Driven Enterprise Real-time Translation SaaS Platform

2024 – Present
Project Lead & Chief Architect
  • Led NestJS microservices with PostgreSQL and Redis, achieved millisecond-level API response
  • Self-developed speech-to-text engine; integrated and optimized DeepSeek AI for translation; designed subscription tiers for commercialization

Malaysia Smart Tourism AI Assistant

2025 – Present
Core Developer & AI Algorithm Lead
  • Fine-tuned Gemini 2.5 Flash on Google Vertex AI for Malaysian context; designed four-stage dialogue engine with multimodal interaction (FastAPI + Streamlit)

AIT Systems — Intelligent Enterprise Management Platform (with Daikin Industries)

2025 – Present
Technical Lead
  • Designed TypeScript-based microservices + RAG stack with Ollama local LLM; RBAC data security; enterprise AI assistant for knowledge retrieval
05

Skills

Programming Languages
Python TypeScript JavaScript Java C SQL
Backend
NestJS FastAPI Django Node.js Socket.IO Microservices
Frontend
React Next.js Redux Toolkit Tailwind CSS Streamlit
Database & Cache
PostgreSQL Supabase TypeORM Redis MongoDB
Deep Learning
TensorFlow PyTorch CUDA TensorRT vLLM DeepSpeed
Computer Vision
DeepFace FasNet RetinaFace MTCNN Emotion Recognition Liveness Detection
NLP & LLM
GPT-4 Claude Gemini DeepSeek Deepgram STT/TTS RAG LoRA/QLoRA/PEFT
AI Frameworks
LangChain LlamaIndex Ollama Function Calling Multi-Agent Systems
Cloud & DevOps
GCP (Vertex AI) AWS Docker Kubernetes CI/CD Git
06

Certifications

1
Python for Everybody Specialization — University of Michigan
2
C for Everyone: Programming Fundamentals — UC Santa Cruz
3
Introduction to Networking — NVIDIA

Writings & Thoughts

Articles on engineering practice, AI tooling, and development philosophy.

Engineering Practice 2025

My AI-Assisted Development Policy

A personal framework for using AI coding tools responsibly — how I stay productive without losing engineering rigor, and the rules I hold myself to.

Read →