Selected Projects
AI Code Review for Azure DevOps
Self-hosted, LLM-powered pull request reviewer for Azure DevOps. A Python CLI and reusable pipeline template review the diff with a model you control and post findings as inline PR comments.
Challenge: Automated reviews tend to drown reviewers in style nits and repeat themselves on every push. The prompt targets only real defects, findings are re-anchored to changed diff lines, hidden markers prevent duplicate comments, and teams can add repository-specific guidance in Markdown. HIGH and CRITICAL findings fail the build, and code only goes to an endpoint in the organisation's own cloud boundary, behind the AI gateway.
RAG Chatbot
Full-stack AI chatbot built with Next.js, the Vercel AI SDK and PostgreSQL with pgvector. It answers questions from its own knowledge base, streams responses and stores chats per user.
Challenge: LLMs guess when they lack facts, and adding knowledge usually needs a separate ingestion workflow. Here the model uses tool calling to decide when to store or retrieve knowledge, so users can teach it new facts in plain conversation and get answers grounded in them.
View on GitHubPython Observability Decorators
Lightweight Python decorators and structured events for data pipelines in Microsoft Fabric, Databricks and PySpark. Events go to Azure Event Hubs or a Fabric Eventstream.
Challenge: When a step in a notebook-based pipeline fails, you often learn about it late and without knowing which run, table or parameters were involved. The decorators attach a run-wide context to every error event without touching existing function bodies, and re-raise the exception so orchestrators still see the failure.
View on GitHubLightweight Data Entry App
Vue 3 and Node.js web app for browsing, editing and validating BigQuery tables, driven by metadata tables rather than hard-coded schemas.
Challenge: Business users need to maintain reference and planning tables, but should not get direct database access. The backend validates every change set against the table metadata, applies it and writes each change to an audit log, so edits stay controlled and traceable.
View on GitHub