Practical AI systems for legacy workflow modernization.

Problem: A group of family finance spreadsheets had become too brittle for transaction imports, credit card cycles, budgets, and reporting.

Built: A family finance application that imports bank transactions, credit card statements, and related documents through a web interface. From there, I can categorize transactions, manage budgets, generate reports, track monthly credit card cycles, and record payments.

Stack / Tools: AWS serverless architecture, Claude API for transaction categorization, developed with Claude Code, modern web frontend.

Project link: GitHub repo coming soon

AI angle: An AI category suggester analyzes uncategorized transactions and proposes new automation rules through a Claude API workflow.

Problem: Strata council operations were buried in a high-volume inbox with no reliable way to organize issues, decisions, and budget relevance.

Built: A council operations application that turns incoming council emails into a structured workflow. Messages can be assigned to topics, grouped into issue threads, marked for budget relevance, and moved toward decision workflows.

Stack / Tools: AWS serverless architecture, email ingestion workflow, topic/message data model, developed with Claude Code, modern web frontend.

Project link: Live site →

AI angle: The product direction is toward structured council intelligence: message classification, topic summaries, decision history, and board-ready context.

Problem: Billing and collections for a consulting business were managed through manual steps, spreadsheets, and email threads with limited cycle visibility.

Built: A billing and collections application that imported invoices, accepted invoice PDFs, extracted key invoice data, and organized everything into monthly billing-cycle tables. The web UI showed sent invoices, status, key data points, and collections progress.

Stack / Tools: GCP serverless architecture, DocumentAI, developed with Google AI Studio, Antigravity, Gemini, web UI for billing-cycle operations.

Project link: Case study only

AI angle: DocumentAI turned invoice PDFs into structured operational data, making reporting and follow-up possible inside the application.

Problem: High-volume inboxes bury important work under newsletters, misrouted messages, accounting emails, and general notifications. The daily triage ritual can consume focus before the workday has really started.

Built: A Claude Cowork skill that reads unread Gmail, categorizes each thread against configurable rules, applies safe Gmail actions, and sends a Slack digest summarizing what was processed.

Stack / Tools: Claude Cowork skill, Gmail connector, custom Gmail modify MCP server, Gmail API, Slack digest, Python, config-driven category rules.

Project link: GitHub repo →

AI angle: The skill combines agent reasoning with deterministic guardrails: configurable categories, safe batched Gmail modifications, and a daily operator-facing summary.