Quill — AI-Powered SOW & Proposal Generator

Year

2026

Client

Nexus Labs (anonymised, per the published case study)

Role

Founder & Lead Engineer, Refactrd — designed and built the full system

Built at

Refactrd

Read Full Case Study

Overview

Quill is an AI proposal and SOW generator built at Refactrd for a production studio that generates multiple proposals a week. It takes pasted discovery-call notes, extracts the key project details with Claude, matches the project to the right service packages from a pricing database, and generates all eight standard SOW sections in a branded, exportable PDF. Every section stays editable before export, and the user can customise the PDF's colour palette, font and size before downloading.

Challenges

Manual proposal writing took two to three hours per proposal and produced inconsistent output, and slow turnaround cost deals when a proposal arrived days after a discovery call. The public demo runs on a representative set of service packages rather than the client's real commercial pricing and client information, to protect confidentiality; the workflow, AI and PDF output are identical to what is used in production.

Solutions

Extraction step: Claude turns pasted discovery notes into structured JSON (client, industry, problem statement, requirements, timeline and budget signals) which the user reviews and can correct before generation, since AI extraction is accurate but not perfectGeneration step: a second Claude call with tool use queries a Supabase pricing-package database to select two or three matching packages, then generates all eight SOW sections (Executive Summary, Scope of Work, Phasing and Timeline, Deliverables, Investment, Exclusions, Assumptions, Payment Terms) in a senior-consultant toneInline section-by-section editing directly on the rendered document, with no separate editorBranding step: the user picks from 10 colour presets and 10 fonts before the backend renders the HTML template with the chosen variables and converts it to a PDF via WeasyPrint

Results

Early testers who normally spend around two hours per proposal reported first drafts landing at roughly 80% of the way there after about 12 minutes of generation, with one proposal sent to a client with only minor edits. A production version customised for a specific business typically takes three to five weeks to build, depending on pricing-structure complexity and CRM/proposal-tool integrations.