Research Scientist and AI Lead working on tabular deep learning, LLM applications, and multi-agent systems.

My work sits between research and engineering: developing models, designing APIs, building reproducible workflows, and turning machine learning ideas into systems that can be tested, deployed, and maintained.

• DeepTab: Core developer and maintainer of DeepTab, an open-source Python library for deep learning on tabular data. It provides PyTorch and Lightning-based models including Mambular, MambaTab, FT-Transformer, AutoInt, ENODE, NDTF, Trompt, TANGOS, and MLP-style baselines etc., all exposed through a scikit-learn-style workflow for training, evaluation, benchmarking, and research.

• PreTab: Core developer and maintainer of PreTab, an open-source Python library for tabular feature representation and preprocessing. PreTab focuses on numerical encodings, spline-based transformations, adaptive feature representations, neural basis expansions, and scikit-learn-compatible pipelines.

• LLM-powered information extraction: Leading a production system for extracting structured information from complex business documents. The system combines OCR, LLMs, FastAPI, MLflow, Docker, MongoDB, and cloud deployment patterns for scalable and reliable document understanding.

• Multi-agent enterprise assistant: Working on a multi-agent assistant that connects LLM agents with enterprise systems such as Salesforce, SAP, internal knowledge sources, web applications, and Microsoft Teams. The system supports business workflows such as question answering, order lookup, order creation, order updates, tool calling, agent orchestration, and agent-to-agent communication.