Here's what I built, wrote, and proved. Across financial strategy, AI engineering, and academic research.
An agentic AI solution for corporate finance
Foresight is an agentic AI solution currently under development that answers the "what if" questions corporate finance teams ask, turning raw data into decision-ready analysis.
Early result: Foresight outperformed senior human corporate finance experts on the blind evaluations. It also reduced analysis time from ~30 hours (human expert) to ~1 hour.
A RAG-Powered AI Agent for Regulatory Compliance — From Weeks to Minutes
A self-developed project that designs an autonomous AI agent to automate the regulatory change lifecycle for financial institutions. A head-to-head comparison (Gemini CLI + RAG vs. ChatGPT Pro vs. Grok 4) showed that:
RAG-grounded analysis catches compliance gaps the other tools miss. Across three use cases, the agent reduced impact analysis from 1–2 weeks of manual review to under 5 minutes.
An Agentic AI Pipeline for Family Spending — From Receipts to Insights
Every receipt, screenshot, and forwarded online order confirmation becomes a structured row (date, store, item, price), so the data is later queryable. Want to know which store is cheaper for a recurring product, how this month's travel spending tracks against budget, or how spending patterns have shifted?
The result:The AI captures the detail; SQL answers the questions. It cut monthly manual processing time from 3 days to under 2 hours.
An Automated Pipeline for Financial Statement Tagging — From Raw Filings to Standardized Data
A Python pipeline that automates the tagging of SEC EDGAR filing data into standardized financial statements across a retail peer set. Since every company labels the same line item differently, a tag dictionary and position-based classification map labels automatically — no hardcoded mapping to maintain, and data is pulled directly from XBRL facts and filing text.
The result: a fully automated, auditable path from raw, inconsistent filings to standardized financial data — no manual tagging or reconciliation required.
Rental Investment Strategy with a Data-Driven Solution
Built a machine-learning model to predict rental prices for U.S. apartments using property features, amenities, and neighborhood income data. Random Forest delivered the best overall performance, while state-specific models captured regional pricing differences.
The result: the model flags listings priced significantly above or below prediction and groups properties into distinct market segments (e.g., luxury, budget, family-sized), helping prioritize opportunities for deeper analysis.
Mid-Level Knowledge Workers Are Facing the Threat of AI Replacement
AI surpasses experienced human professionals on two dimensions: speed (solutions within minutes) and depth (more persuasive reasoning and theoretical grounding).
Powered by an original Multi-AI Cross-Validation design: Gemini and Claude as analysts, ChatGPT as examiner. Written in 3 days and 20 hours from concept to completion.