Hello, I'm
Research Scientist at Salesforce AI Research, working on time series and tabular foundation models, from research to production. PhD in CS from NUS, with a focus on foundation models, agentic systems, and resource-efficient generative AI.
I am a Research Scientist at Salesforce AI Research in Singapore, where I work on research and production for time series and tabular foundation models. I specialize in foundation models, agentic systems, and resource-efficient generative AI (LLMs and VLMs), and I enjoy taking cutting-edge research from zero to one and into enterprise production.
Before Salesforce, I was an AI Scientist at SAP in Singapore, where I contributed to training a document foundation model, fine-tuned and deployed lightweight Vision-Language Models for agentic document processing, built LLM agents and RAG systems for the SuccessFactors platform, and led AI safety work on guardrails against jailbreaking and bias mitigation.
I completed my Ph.D. in Computer Science at the National University of Singapore as a member of the Embedded Computing Lab (EcoLab), supervised by Prof. Tulika Mitra (NUS) and co-supervised by Liu Siying and Pham Nam Trung from A*STAR. My research focused on efficient deep learning — particularly data-free knowledge distillation, model compression, and generative pseudo replay. My PhD was supported by the prestigious Singapore International Graduate Award (SINGA).
I hold a B.S. in Computer Engineering from Bilkent University, Turkey, where I graduated with High Honors and the Data Science Award. My work has appeared at top venues including AAAI, EMNLP, RA-L, and WACV.
Outside of work, I enjoy 💃 Argentine tango, 🏋️ gym, 🍸 cocktail making, ☕ coffee brewing, ✈️ travelling, and 🧗 bouldering.



Selected peer-reviewed publications at top-tier venues
A real-time AI game engine that turns mystery novels into interactive whodunit games, with autonomous AI agents playing the characters and guiding players. An end-to-end pipeline generates custom mystery experiences from short text prompts.
Finalist — Mistral AI Hackathon 2026
An autonomous travel-planning agent built on LangGraph that combines Google Maps, weather APIs, and real-time web search to build dynamic, context-aware itineraries, with long-horizon state and context management for multi-turn planning sessions.
An AI mobile assistant for driver safety with real-time traffic sign detection, gaze monitoring, and collision risk prediction. RFCN models were trained and quantized from TensorFlow to TFLite for low-latency on-device use. Senior design project; led a team of 5 engineers.
Data Science Award — Bilkent University