Problem
Ad-hoc call notes are hard to convert into consistent end-of-day updates without manual cleanup and formatting overhead.
Approach
- Designed a serverless stack focused on transforming raw notes into structured reports.
- Used Cloudflare Workers patterns and lightweight API-first service design.
- Framed the product as an applied AI workflow rather than a generic note tool.
Architecture snapshot
Edge runtime: Cloudflare Workers
Workflow: Notes ingest -> transform -> summary output
Product lens: Operator-first reporting workflow
Outcome
- Live product narrative on Base Layer Labs with clear production use-case positioning.
- Strong bridge between product engineering and applied data/AI workflows.
- Fast iteration cycle with low operational overhead.
Artifacts and proof
Demo video
Loom walkthrough (~90s): Record using the script in the portfolio repo (docs/LOOM-GHOSTSTACK-SCRIPT.md), then paste the Loom share URL on the line below — Notion will embed the player automatically.
- Repository: davidlwilson2021/ghoststack
- Product surface: baselayerlabs.dev/product
- Positioning context: baselayerlabs.dev
- Recording script (repo): LOOM-GHOSTSTACK-SCRIPT.md
// Workflow contract pattern used in the project
type DailyReportInput = {
accountName: string;
callNotes: string;
dateISO: string;
};What I learned
- Product clarity improves when user workflow language is explicit in the interface and copy.
- Operational simplicity matters as much as model sophistication.
- Serverless architecture keeps iteration cycles fast for solo teams.