Singapore, Singapore
From driving the yield ramp for the industry's first 176-layer 3D NAND to pushing the boundaries of manufacturing AI. I recently architected our flagship code-first GenAI framework. Anticipating the agentic "skills" paradigm, I encoded our domain as deterministic code using hexagonal architecture and DDD. Powering flagship frontend, backend, and cost agents, this foundation delivers near-100% accuracy, reduces LLM latency by >3x, and saves hundreds of thousands of engineering hours annually.
Architected our flagship code-first genAI framework. Anticipating the agentic skills paradigm, I encoded complex manufacturing logic into deterministic, executable code via hexagonal architecture and domain-driven design. Engineered LLM expert systems for automated root-cause analysis. Pioneered a multi-dimensional evaluation rubric and utilized chain-of-thought/code-assisted execution to boost LLM accuracy on complex issues from ~0% to 60% (>95% for standard cases). Architected an adaptive, role-based Python data-frame library using dynamic structural typing and C-level memory optimizations. Processing terabyte-scale trace data interactively, it slashes memory usage by 10x, boosts speeds up to 75x, and cuts analysis time from weeks to days. Optimized a C++ diagnostic application operating on >20 billion geometric shapes. Reduced the core algorithm's time complexity from quadratic to logarithmic, cutting processing time by a factor of 8,700. Engineered significant data pipeline optimizations, including rewriting a MapReduce pipeline to reduce execution time from 8 hours down to 5 minutes. Built widely adopted manufacturing apps, including the company’s #1 ranked material tracking web-app, a Python-based fail-map diagnostic tool, and Java Spring microservices for an enterprise Analytics Suite.
Was responsible for conceptualizing new ideas to improve performance and yield of 539 CMOS and 37 NAND Replacement Gate manufacturing steps, and executing said ideas in a manner safe for mass production Enhanced transistor electrical performance through modification of semiconductor doping, resolved top issue causing defective dies with short circuits and reduced breakage of NAND chips by implementing a new manufacturing diffusion process Was promoted from Engineer Level 1 to Engineer Level 2 ahead of peers due to good performance and provision of valuable insights
Coordinated amongst schedulers, refinery coordinators and marine transport optimizers to minimize shipping costs Proactively found opportunity to save USD 900 000 annually through data analysis of berth occupancy Performed analysis that triggered project to de-bottleneck post-pump activities, projected to save USD 600 000 annually
Worked with supply department to propose a supply chain (procurement and production) strategy for SOPEO’s packaging plant, with further plans for using convex optimization via model predictive control, predicted to reduce costs by 11% Produced a condensed 31-page report for reducing utilities and raw material usage, major cost drivers in SOPEO refinery, with recommendations for use of response surface methodology, evolutionary operation, and ARX/ARMAX modelling Presented recommendations to management and received approval for implementation
Developed five applications for use by fund managers, completing five projects in length of time assigned for 2 projects Designed a credit model application, completing the design in three days instead of the allotted 30 days Developed automated three-dimensional visualizations of attributes of fixed income securities and an automated system to alert the Fixed Income team in the event of a fund-mandated credit rating or country exposure limit breach