A production project management tool for PE operators — built autonomously in two overnight missions. 26,000+ lines. Zero developers. Zero human-written code.
Every project management tool is built for the wrong audience. Jira is for dev teams. Kantata is for finance admins. Monday is for everyone and therefore no one. None of them are built for the operator — the person logging three minutes between calls, turning a transcript into five action items, assembling a steerco from work that's already been done.
The stakes are high. In Nordic PE, reputation is everything. One bad first impression with amateur tooling and doors close across firms. The project management tool is visible to the partners who decide whether to engage you again.
Two missions. One night each. A complete application.
All engagements at a glance. Status, health, next actions. The view an operator checks every morning.
11 tabbed modules per engagement: overview, workstreams, actions, risks, decisions, stakeholders, time, invoices, files, transcripts, Gantt.
The signature interaction. Log time in 5 seconds. Keyboard-first. Workstream pre-fill from context. The thing that makes an operator use this daily.
Rate-aware roll-up from time entries. Inline line editor. Print-ready HTML preview with tokenized bank details. Approval workflow.
Separate route group — completely isolated from internal views. Project health, decisions (approve/reject), deliverables, invoices. What a PE partner sees.
Upload meeting transcripts. AI extracts action items, decisions, risks. Side-by-side review: raw transcript + extracted items. One-click save.
Assembles steering committee materials from work already done. AI polish with claim-matching (anti-fabrication). Diff against prior steerco.
RLS on every table. PII tokenization. Hash-chained audit log. GDPR Article-level mapping. DORA-readiness phased honestly. Built in, not bolted on.
| Metric | Mission 1 | Mission 2 | Total |
|---|---|---|---|
| Build time | ~90 min | ~111 min | ~3 hours |
| Tickets | 10 | 43 + 5 | 58 |
| Files | 51 | 154 | 205 |
| Lines of code | 3,398 | 23,271 | 26,669 |
| Database tables | 17 | — | 17 |
| RLS policies | 33 | — | 33 |
| SQL migrations | 9 | — | 9 |
| API routes | 12 | 18 | 30 |
| UI components | 0 | 50 | 50 |
| App routes/pages | 0 | 59 | 59 |
| TypeScript errors | 0 | 0 | 0 |
| Human code written | 0 | 0 | 0 |
13.7x amplification. A 992-line mission spec produced 13,650 lines of production UI code. One page of intent, one working application.
| Layer | Technology |
|---|---|
| Frontend | Next.js 15 App Router, React 19, TypeScript, Tailwind |
| Backend | Next.js API routes, Supabase client |
| Database | Supabase (PostgreSQL 17), 17 tables, RLS, pgsodium tokenization |
| Auth | Supabase Auth, SAML scaffolding, ABAC with tenant-scoped predicates |
| Architecture | Hexagonal (ports & adapters), 6 interfaces, provider-swappable |
| Audit | SHA-256 hash-chained append-only log, tamper-evident |
| Compliance | GDPR (Art. 17/28/30/32), DORA-readiness, PII tokenization |
This isn't a demo. It's a production application — with tenant isolation, PII tokenization, compliance documentation, and a print-ready invoice system. Built in the time it takes to watch a movie. Twice.
The question isn't whether AI can build software. It can. The question is whether it can build software you'd trust with your clients' data, your firm's reputation, and your own professional credibility.
26,669 lines. 33 RLS policies. Hash-chained audit trail. Zero human code. The answer is yes — when the build is governed.
Application screenshots will be added after deployment and visual review.