AI-powered quote automation that cuts proforma prep from ~30 minutes to ~15 seconds.
A multi-tenant B2B SaaS platform for manufacturers.
The Problem
Manufacturers build proformas in Excel or Word — looking up product codes, copying prices from a list, applying discounts and VAT manually. Every order is a mini data-entry project.
Wrong product codes, stale prices, and manual math mistakes slip into customer-facing documents — damaging trust and requiring costly re-work or re-sending.
A sales team handling a steady flow of orders can spend most of its time on data entry instead of selling — growth bottlenecked by process, not demand.
The Solution
From a raw customer order to a signed-off PDF proforma — fully automated.
A sales rep types or pastes a customer order in plain language — WhatsApp message, email snippet, or handwritten note. Any format accepted.
OpenAI GPT-4o-mini extracts customer name, product codes, quantities, and discounts. Output is validated by Zod — never free text into the database.
Each code is matched against the company's catalog by normalized_code. Unknown products surface a correction UI — prices are never invented.
Line totals, group discounts, and VAT are computed deterministically from catalog data. Zero manual math, zero price errors.
A branded, ready-to-send proforma PDF is generated via @react-pdf/renderer and delivered — in ~15 seconds total from order entry.
Illustration — generic placeholder data only
Customer: Sample Customer Co.
Product A — qty 10, 40% disc
Product B — qty 5, 30% disc
Product C — qty 20
How I Built It
Quotivo is built on a Next.js 16 App Router backend-for-frontend with API routes handling all server-side logic. Supabase provides authentication, a managed Postgres database with Row-Level Security, and object storage for PDF assets. The AI parse layer calls OpenAI GPT-4o-mini and all model output is validated against a Zod schema before any database write. PDF generation runs via @react-pdf/renderer in a non-blocking, fire-and-forget job.
Every Postgres query is scoped by company_id, enforced at application layer and database layer via RLS. Tenants are completely isolated — no cross-tenant data leakage possible.
The model never returns free text into the database. All parse output goes through a strict Zod schema. Unknown products surface a correction UI, never invented prices.
PDF generation and Resend email delivery are kicked off after the API responds. Users never wait for I/O-heavy jobs, keeping the interface responsive.
The Build Process
Quotivo was built without a traditional software engineering background, using an AI-assisted, prompt-driven development workflow — including Claude Code for agentic code generation and review. This isn't a story about learning to write code; it's about applying strong product thinking and architectural judgment to direct an AI development pipeline toward a real business outcome.
Every security decision — RLS policies, tenant isolation, schema-validated AI output, service-role client restrictions — was deliberately designed, audited, and documented. The codebase includes a formal security audit report, a risk matrix, and architecture decision records.
The result is a production-deployed, multi-tenant SaaS with a paying pilot customer — built by understanding what to build and why, and using AI as a high-leverage implementation tool.
Results
Pilot: a mid-sized furniture manufacturer (kept anonymous). Tested on their real product catalog and live order volume.
Tech Stack
I'm available for freelance projects, SaaS consulting, or full-stack development contracts. Let's talk.