Computer Engineering @ NTU · ML Systems · GenAI · Quant Finance · Full-Stack

I build production-grade ML systems, quantitative backends, and multimodal AI pipelines. Currently interning at Panasonic R&D Centre (ML & GenAI Engineer) and incoming Data Science Intern at Vertex Holdings.

PyTorch HuggingFace TRL vLLM Triton Qdrant FastAPI Docker

• Fine-tuned Qwen-2.5-7B-Instruct on code generation using QLoRA (4-bit NF4) with GRPO & DPO post-training alignment
• Implemented LoRA from scratch in NumPy/PyTorch (W = W₀ + αBA/r) with manual backprop
• Authored Triton fused softmax kernel reducing HBM passes 3→2; benchmarked HumanEval pass@1 & MBPP
• Built end-to-end RAG pipeline (bge-small-en-v1.5 + Qdrant + MMR retrieval); deployed vLLM + FastAPI on Docker Compose with Prometheus TTFT/P95 monitoring

LangGraph Qwen-2.5-7B vLLM RAGAS mem0 Qdrant Celery Redis Prometheus Grafana

• Engineered a 5-agent LangGraph StateGraph with human-in-the-loop checkpointing and conditional routing
• A/B tested zero-shot vs CoT vs few-shot across 20 benchmark queries — CoT achieved RAGAS faithfulness 0.83 (+0.12 over zero-shot)
• Deployed async production serving via FastAPI + Celery + Redis (30–120s job queue); 94% citation accuracy, 88% tool-selection accuracy
• Instrumented per-agent Prometheus + Grafana P95 histograms with GPT-4o-mini LLM-as-judge evaluation

PyTorch CLIP ViT-L/14 Whisper DeepFM MMoE FAISS HuggingFace Celery C++

• Fused CLIP ViT-L/14 visual embeddings (mean+max temporal pooling → 1536-dim) with Whisper-base ASR transcripts; late-fusion MLP achieved mAP 0.67 on 20-class topic tagging and F1-macro 0.74 on safety filtering
• Deployed two-stage pipeline: two-tower retrieval (InfoNCE, FAISS) + DeepFM ranker (AUC 0.77) + MMoE multi-task scorer (AUC 0.73, watch-time/like/share gating)
• Implemented MMR re-ranking (λ=0.3) with FAISS cold-start fallback (NDCG@10 +0.19 delta)
• Accelerated frame extraction 3.1× via C++/OpenCV pybind11 binding

Python FastAPI Apache Kafka TimescaleDB Redis Docker

• Production quant engine: Monte Carlo VaR (10k paths), CVaR/Expected Shortfall, Markowitz Efficient Frontier via SLSQP optimisation
• Event-driven market data pipeline via Kafka (KRaft) → TimescaleDB with Redis caching at <5ms hit latency
• Fama-French 3-Factor OLS regression with rolling 126-day factor exposures and CCAR-style stress tests (2008 GFC, COVID, rate shock, stagflation)

TypeScript Node.js WebSockets Swagger React Vite

🥈 2nd Place — Hacx! Hackathon (HTX × Microsoft)

• Built a TypeScript/Express backend with modular controllers, services, and Swagger documentation for transport monitoring across 15+ endpoints
• Implemented deterministic telematics and vital-sign simulation with threshold-based alerting for critical safety events
• Designed audit and session tracking workflows with structured logging and in-memory state management

Solidity Ethers.js Hardhat Flask MongoDB

🌏 Global 5th Place — APRU × Google Tech Policy Hackathon (Bangkok)

• Built Flask REST backend supporting marketplace, futures, contract, and insurance workflows with MongoDB persistence
• Integrated backend services with Solidity smart contracts and TypeChain bindings for auditable on-chain/off-chain settlements

TypeScript Node.js React SQL Prisma ORM OneMap API

Academic Project — Software Engineering (SC2006)

• Built Node.js/Express backend with REST APIs for property discovery, geospatial search, and favourites
• Engineered ETL pipelines to ingest, geocode, and normalise large public datasets into analytics-ready MySQL tables via Prisma ORM

PyTorch Node.js FastAPI PostgreSQL Docker

🥉 3rd Place — Deep Learning Week Hackathon (NTU × Jane Street × Microsoft × OpenAI)

• Built a containerised Node.js backend exposing REST APIs for real-time governance and risk evaluation
• Implemented a policy + ML fusion engine combining rule-based checks with PyTorch outputs to generate auditable risk verdicts

Languages Python TypeScript C++ Java C# SQL Solidity CUDA

ML / AI PyTorch HuggingFace (TRL, PEFT, Transformers) LangGraph LangChain vLLM FAISS Qdrant Triton W&B RAGAS

Backend & Infra FastAPI Node.js Express Celery Redis Apache Kafka TimescaleDB PostgreSQL Docker REST API WebSocket

Frontend React Vite Next.js

Data & Analysis Pandas NumPy Matplotlib

Nanyang Technological University (NTU), Singapore — Aug 2024 – May 2028 Bachelor of Engineering (Honours), Computer Engineering

Relevant Coursework: Data Structures & Algorithms, Linear Algebra, Probability & Statistics, Discrete Mathematics, Software Engineering, Computer Networks, OOP, Algorithm Design & Analysis