Experience

AI Engineer

General Magick Industries

May 2026 – Present

Re-engineered a multi-model LLM reasoning service, optimizing the Monte Carlo Tree Search (MCTS) pipeline to achieve ~5.3× faster execution than the previous implementation through parallelized evaluation and aggregation.
Benchmarked LLM reasoning strategies using datasets such as GPQA, assessing answer quality, latency, and inference cost to inform model routing and selection, with successful configurations completing in ~36 seconds.
Extended MagickMind’s open-source Python SDK with multi-model reasoning support, enabling developers and internal services to access these capabilities through the platform API.
Co-designed and implemented agent memory architecture based on the CoALA cognitive architecture, covering semantic, episodic, and procedural memory with consolidation, forgetting, and memory-decay mechanisms.
Owned implementation of usage metering and billing infrastructure, including usage tracking, wallet-credit management, invoice validation, and duplicate-event handling.

PropTech Builders

Jul 2025 – May 2026

Previously Data Scientist (Intern)Apr 2025 – Jun 2025

Owned the processing pipeline for 2M+ raw property listings, turning inconsistent portal data into standardized, enriched records for valuation and reporting.
Designed cross-portal property matching using property names and coordinates, followed by unit details such as floor, area, and room counts. Applied CLIP-assisted image similarity to help identify listings of the same unit.
Eliminated 87% of identified duplicate listings, with manual checks for false matches. Linked source listings to a canonical property record and combined their information to fill missing attributes.
Monitored data quality, prepared inputs for automated valuation, and automated explainable valuation reports for brokers and internal teams.
Worked with founders and product owners to turn ambiguous property-data requirements into implemented processing and reporting workflows.

Contributed to a Myanmar-language humor dataset, annotating long-form text across four categories to support low-resource NLP research.
Fine-tuned transformer models, including BERT and mT5, for humor classification on Myanmar-language text.
Applied back-translation and paraphrasing to expand the training data and address limited annotated-language resources.

Built a Django maintenance-reporting application with issue routing, dashboards, and visualizations to help factory teams track operational issues.
Implemented data-processing workflows to automate manual data-entry tasks for the production team.
Collaborated with operations and engineering teams to gather requirements and deploy improvements to their reporting workflows.

King Mongkut’s Institute of Technology Ladkrabang (KMITL)

Led tutorial sessions and guided students through assignments and exam preparation across software engineering, AI, electronics, and logic courses.
Assisted instructors in facilitating lab sessions and supported students with course exercises.