Projects I have built

A repository of deep learning projects to teach me how to train models from scratch (or from pre-trained models) using PyTorch and the Fast.ai book.

Clusters customer feedback at scale, surfaces product-level themes and blind spots, and improves insight quality through evaluation loops.

Generates structured test suites for AI agents (golden path, edge case, and adversarial tests) from a system prompt or GitHub repo URL. Surfaces how an agent behaves when things go right, when assumptions break, and when it is actively pushed off course.

Open-source agent workflow that turns essays into X posts, routes review through Google Sheets, posts approved content automatically, and learns from engagement data.

Captures product feedback directly from Slack reactions, prepares it for ingestion into a classification pipeline, stores it for review, and exports approved insights as structured CSV.

Slack app that turns a thread into a PRD draft, thread summary, or action items using slash commands and OpenAI, with post-to-thread and regenerate flows.

Multi-user SaaS that turns a sales deck PDF into 10 LinkedIn post candidates in under 90 seconds, written in your voice. Review and approve in Google Sheets, publish directly to LinkedIn. Currently in private beta.

Multi-step inbound lead tool: visitors answer structured questions about their workflows, Claude generates a personalised automation quick take, and the full assessment lands in my inbox via Resend.

Interactive experience for user-owned memory that stays private by design, with permission selection to grant AI apps access only to specific aspects of personal context.