Acc. XIV · Chartula reticulata — branded document engine, cultivated 2026
Resibo — AI quote, invoice, and receipt maker
Technical founder · Product engineer · 2026 · Independent studio
An AI document generator for Philippine micro-businesses that turns a rough Taglish description into a branded quote, invoice, or acknowledgment receipt.
Chartula reticulata — branded document engine, cultivated 2026
Field notes
The context
Micro-businesses in the Philippines write quotes and receipts by hand or in loose chat messages, with no consistent branding, no payment tracking, and no reusable customer record. Resibo takes a rough Taglish description and drafts a branded document, then tracks the payment against it.
Cultivation
What I built
- Built AI drafting from Taglish input through a Groq/Cloudflare/OpenRouter fallback chain, generating quotes, invoices, and acknowledgment receipts.
- Built image and PDF export with seller branding, plus payment tracking that supports partial payments.
- Built the customer directory, auto-accumulated from generated documents rather than entered separately.
- Built the offline-first PWA sync queue with conflict resolution, and tier-based feature gating enforced server-side against subscription state.
- Built the 14-day trial with manual payment verification through Mission Control, and the admin console for tenant lifecycle, payment ops, and audit logging.
- Wrote 343 test files across Vitest and Playwright covering the core drafting loop, offline sync, subscription lifecycle, admin actions, and the landing page.
Root system
How it works
A Turborepo with three independently deployed Vercel apps: @resibo/landing (Next.js), @resibo/app (a Vite PWA), and @resibo/mission-control (a Vite admin app), sharing @resibo/schemas (Zod), a /ui package holding both the design system and the document renderer, /i18n, and /config. The Supabase backend spans 13 migrations — enums, RLS policies, stored procedures, Edge Functions (llm-proxy for drafting, approve-payment for subscription ops), pg_cron for lifecycle reconciliation, and audit triggers. Auth is two-tier: customers via email/Google, admins via a super-admin flag, with every write validated server-side by RLS policies checking tenant status and paid_until.
Harvest
Outcomes
- MVP shipped; all three surfaces — landing, app, Mission Control — are live.
- No users or revenue tracked to date.
- Payment verification is manual with a human review SLA; there is no payment gateway.
- LLM inference runs on free-tier providers at MVP scale, which caps throughput until that changes.
Herbarium
The work itself






A fallback chain across three LLM providers turned out to matter more than picking the 'best' one — free-tier rate limits and outages are the actual failure mode at this scale, not model quality. Deriving the customer directory from documents rather than a separate CRUD flow removed an entire onboarding step users would have skipped.