Machine Learning Engineer

• Fine-tuning at scale (LLaMA-3.1 70B): Built an end-to-end pipeline with PEFT (LoRA/QLoRA) + RAG over internal domain data, improving task accuracy on internal benchmarks by 25% and reducing production drift by 15% over the evaluation window; governed by offline holdout tests and progressive traffic gating.
• Agentic AI (orchestration & planning): Integrated a multi-agent workflow directly into the same application. A planner/router decomposes user intents into sub-goals, selects tools (internal search, scraping, structured DB lookups) based on a question taxonomy, and emits step-by-step CoT plans to guide execution.
• Agentic AI (evidence retrieval & verification): Implemented retrieval/scrape agents for structured and unstructured sources with source-level citation tracking, plus a verification agent that runs CoT-based cross-checks. This improved factual grounding vs. a single-agent baseline by 18% while keeping response times within the application SLA through caching and bounded tool-use.
• High-performance inference & deployment: Productionized with vLLM on AKS across A100/H100 nodes; GPU memory/KV cache optimizations cut p95 latency by 40% and lifted throughput by 50%, reliably serving 10K+ daily requests under peak load.
• Integrated robust CI/CD processes, monitoring, and automated scaling strategies to ensure continuous model improvement and reliable production deployments across cloud-based environments.
• Architected scalable data pipelines: Implemented robust data processing solutions using SQL Server, Azure Databricks, and PySpark, cutting processing times by 30% and significantly enhancing overall system performance.
• Integrated CI/CD, monitoring, and automated scaling: Established end-to-end processes that ensured continuous model improvement and reliable production deployments across cloud-based environments.

NLP Research Intern

• Developed a machine learning language model tailored for English-Malay code-switched data, achieving a 20% improvement in accuracy over baseline models by implementing advanced statistical and neural augmentation techniques.
• Integrated linguistically informed algorithms—including part-of-speech tagging and grammatical coherence—to enhance multilingual NLP robustness and advance the state-of-the-art in code-switching language processing.
• Enhanced the model's ability to handle diverse linguistic patterns, advancing the state-of-the-art in code-switching language processing.
• Contributed to the development of bilingual communication technologies by applying cutting-edge machine learning techniques for code-switching scenarios.

Bachelors of Engineering

ACL ARR December 2024