Loading…
Functional, deployed MVP
NAK President Command Center
An AI-assisted operations dashboard built with Claude Code to organize presidential responsibilities, documents, deadlines, meetings, events, and officer follow-ups.
Organization
Personal project, built for Nu Alpha Kappa Theta Chapter presidency
Role
Product Owner and AI-Assisted Developer
Period
2026
Status
Functional, deployed MVP
Live and deployed. Typecheck, lint, and the full automated test suite (87/87) pass; the production build succeeds across 60+ routes. After launch, a real chapter meeting transcript surfaced an actual bug, which was diagnosed and fixed, not just tested in the abstract.
The problem
As chapter president, the information I needed to track was scattered everywhere: chapter documents, officer responsibilities, deadlines, national and regional requirements, screenshots and announcements, meeting notes, and follow-up items spread across group chats, email, and my own notes. I wanted one private dashboard that could help me stay on top of it and reduce the odds of missing a deadline or a follow-up.
The goal
Build a working system, using AI-assisted software development, that would actually reduce what I had to hold in my head as president, without spending months learning full-stack development first.
My role
I was the product owner and the AI-assisted developer, not a hands-off requester. I identified the problem, defined the product requirements, organized the source documents (bylaws, policies, meeting records) so Claude Code could work from them, directed the application architecture and database design through detailed prompts, reviewed what was generated instead of accepting it blindly, worked through Node.js/npm and Supabase setup myself, troubleshot real PostgreSQL and migration errors, tested the app as it was built, and made the call to cut the project's scope back down when it grew past the actual problem I was solving.
Product judgment
The scope pivot
Original concept
The original idea was a broad NAK Chapter Command Center meant to manage members, attendance, dues, fines, strikes, appeals, meetings, chapter governance, officer permissions, and financial workflows: essentially the whole chapter's administration.
Why it changed
Once that broader version was built out, I realized I'd built more than I actually needed to do my own job as president, and that the added complexity wasn't helping me stay on top of my actual day-to-day responsibilities. I made the call to narrow it.
Revised solution
The current NAK President Command Center is a private, single-user dashboard focused on my own responsibilities: tasks, a calendar of events, officer follow-ups, an upload inbox for documents and screenshots, a document library, and meeting/transcript analysis. The original chapter-management system wasn't deleted. It's archived and still fully reachable, just no longer the default view.
Process
- 1Started from a real, recurring problem I had as chapter president, not a hypothetical.
- 2Gathered the chapter's actual bylaws, policies, and meeting records so the AI-assisted build could work from real source material instead of guesses.
- 3Wrote a detailed master prompt describing the application, then used Claude Code to audit requirements and generate the software foundation.
- 4Set up my own local development environment (Node.js, npm) and worked through Supabase configuration, SQL migrations, and database triggers, including real errors that came up along the way.
- 5Reviewed the generated code and database design rather than accepting it as-is, and tested functionality as it was built.
- 6Partway through, reviewed the growing application and recognized the scope had moved past the problem I actually had.
- 7Rewrote the product requirements around a narrower, personal president dashboard, and directed the rebuild.
- 8Designed the document-intelligence and meeting-transcript workflow (including Fireflies.ai exports) with a human review step before anything becomes a real task or event.
- 9Kept testing and refining through further prompts after the core build, including fixing a bug found through a real transcript upload after launch.
Feature status
What's built vs. what's still outstanding
Completed and shipped
- Home dashboard: overdue/due-today tasks, upcoming events, outstanding follow-ups, and items awaiting review
- Tasks: full create/edit/complete workflow with status, priority, category, and due dates
- Calendar: month view plus an upcoming list, linked back to the source document or meeting where relevant
- Follow-Ups: tracking who I'm waiting on (officers, national or regional contacts) without requiring them to have an account
- Upload Inbox: drop in a flyer, screenshot, email, or document, or paste text, for AI-assisted analysis
- Review Queue: every AI suggestion from an upload or meeting waits here for a human decision (edit, approve, reject, or merge); nothing is created automatically
- Document Library: searchable archive of everything uploaded, with extracted text, tags, and a trail back to whatever tasks or events it produced
- Meetings: log a meeting and upload or paste a transcript (Fireflies.ai exports work directly) for a structured summary, decisions, action items, and follow-ups
- Archive: the original chapter-management system (roster, attendance, fines/strikes, dues, votes) is fully intact and reachable, just moved out of primary navigation
Known limitations
- Single-primary-user by design: no multi-officer account system yet
- No Google Calendar sync yet
- No Slack or automatic email reminders
- Officer/follow-up assignees are tracked by name, not a linked account, by design
Technologies
Only what’s actually in the codebase. Nothing added because it was discussed but not used.
Claude CodeNext.jsTypeScriptSupabasePostgreSQLTailwind CSSZodVitestGitVercel
Challenges
- Translating written chapter policy into concrete software requirements without letting outdated documents dictate the design.
- Recognizing when the first version's scope had grown past the actual problem, and being willing to cut it back rather than keep building.
- Working through real Supabase/PostgreSQL migration and trigger errors, not just reading about them.
- Designing a review-before-create workflow so AI-extracted deadlines and responsibilities are never trusted automatically, including explicitly flagging inferred (relative) dates like "next Friday" for confirmation rather than guessing.
- Protecting private chapter documents and meeting transcripts: only extracted text, not files, is sent to the AI provider for documents; images are sent only when actively being analyzed; nothing sensitive is logged.
- A real post-launch bug: after processing a real chapter meeting transcript, the app reported "ready for review" with nothing actually found. I traced it to the AI response being requested in a shape that occasionally broke, split the request into two more reliable calls, and fixed the underlying issue so failures are now reported honestly instead of silently.
Deliverables
- A deployed, private operations dashboard covering tasks, calendar, follow-ups, document intake, and meeting analysis
- An AI-assisted document/screenshot intake and human-review workflow
- A Fireflies.ai meeting-transcript analysis workflow
- A full architectural decision log and testing/privacy documentation for the build
Outcome
The result is a working, deployed dashboard I actually use: tasks, calendar, follow-ups, document intake, and meeting-transcript analysis, all built around a review-before-create rule so the AI never writes to my dashboard without my approval. Beyond the tool itself, I came out with hands-on experience directing an AI coding agent through a real build: writing requirements, reviewing generated architecture and database design, troubleshooting real errors, and making the product call to cut scope when the first version outgrew the problem it was meant to solve.
What I learned
- AI tools still need clear direction: what Claude Code produced tracked directly with how clear and specific my requirements were.
- Generated code and database design still need to be reviewed, not accepted on faith: several real issues only surfaced because I tested the app myself and read what was generated.
- Bigger isn't better. The most useful version of this project was the narrower one, and recognizing that took stepping back and reviewing my own scope, not adding more features.
Skills demonstrated
AI-assisted software developmentClaude CodePrompt engineeringProduct management & requirements gatheringScope managementHuman-in-the-loop AI designDatabase concepts (Supabase/PostgreSQL/SQL)TroubleshootingPrivacy-aware product designDocumentation
Visual evidence
Member names and one financial detail were redacted from these screenshots to protect chapter privacy. Everything else shown is the live application.
Want to talk through how this applies to your team?