Portfolio

2026

Most BI tools make it easy to connect data, build dashboards, and ask questions about it. But when an LLM is introduced into the workflow, a different problem appears: how much access should the model actually have to the underlying data? DataMind is a conversational BI platform built around that question. The core idea is simple: the LLM should only have as much access to the database as the user allows, and even with full access, it can only execute harmless, read-only queries. It is a modular-monolith platform letting users ask questions in plain language over relational databases and get back an auditable answer, table, and chart, with an AST-based SQL guard, a configurable disclosure policy, and a semantic layer securing every model-generated query. Solo build spanning a FastAPI backend, React/TypeScript front end, and a multi-database execution layer.

Dec 2025 – Present

Built an AI system that tailors a LaTeX resume and cover letter to any job description end to end: a two-pass LLM extraction step mines job requirements, ranks experience by relevance, and regenerates every section under hard length constraints so the result always compiles to a fixed page budget. Solo project, designed and built end to end — architecture, LLM prompt engineering, LaTeX generation, local API server, and Chrome extension.

2024

Engineered an end-to-end data pipeline to scrape, normalize, and analyze real-time rental listings, turning unstructured web content into longitudinal market insight on listing duration and regional density. A data project with the City of Ottawa via Algonquin College; I built the scraper, entity-resolution logic, and analysis layer.