AI Researcher, Thoughtworks

Remote | 09/2025 – Present

As part of the AI research team, I am developing a multi-table Q&A data generation framework using LLM-guided SQL-template synthesis and GRPO-based reinforcement learning to improve reasoning and data diversity across complex relational databases.

Key Skills: Reinforcement Learning, Data Generation, SQL Synthesis, Multi-Table Reasoning
Technologies: GRPO, Hugging Face Transformers, LangChain, Python, PyTorch, pandas

ML Researcher, International Finance Corporation (World Bank)

Washington, D.C. | 06/2025 – 09/2025

Led the development of a multi-agent (LLM) system to extract structured insights from semi-structured financial documents such as loan memos and investor reports. Implemented a graph-based traversal framework to identify and tag tables, trace linked notes, and apply model-guided explosion logic for accurate and explainable parsing reducing manual document review time.

Additionally, designed and deployed an Early Warning System (EWS) to detect projects with unusually long processing times in the Mandate-to-Disbursement cycle, delivering predictive insights that enabled proactive regional interventions and helped reduce the median timeline from 286 days toward the management target of 228 days.

Key Skills: Statistical Modeling, Data Analysis, Cross-Validation, Predictive Modeling, Multi-Agent Systems
Technologies: LoRA, LangChain, Azure OCR, pandas, spaCy, Hugging Face Transformers