Wei Xuan Phor

Built and maintained FastAPI-based backend services exposing internal platform capabilities via REST APIs to support MI300+ hardware validation and AI-assisted triage workflows. Integrated AI summarization services via API calls to automate log analysis. Side project: implemented a self-hosted inference service (vLLM) exposed as an API to summarize error logs.

Designed on-premise enterprise LLM chatbot solution (LlamaIndex). Delivered 2 product releases for external stakeholders. Built SQL-retrieval pipelines to query multi-source structured and unstructured data. Benchmarked inference pipelines across 9 LLMs; evaluated 3 RAG strategies; conducted adversarial prompt stress-testing. Integrated Whisper video-to-text; demonstrated multimodal feasibility for internal knowledge search. Improved reliability of an Azure-hosted LangChain web application through LLM output tuning and document refinement. Automated quantization and format conversion (ONNX, PyTorch); cut validation time by 50% across 8 Intel hardware platforms. Performed profiling of AI inference pipelines to identify hardware bottlenecks. Produced Power BI reports on hardware throughput, latency, and CPU/GPU usage.

Improved time-series forecasting model accuracy by 10% using XGBoost and Prophet.