SaaS / AI

Formwise

An AI-powered form builder where you describe what you need in plain English and get a production-ready form with validation, conditional logic, and submissions dashboard.

Full-Stack DeveloperFormwise2024Team of 2
📋

Overview

Formwise lets anyone create forms by describing them in natural language. Type 'job application form with resume upload, years of experience, and salary expectations' and get a working form with proper input types, validation rules, and conditional logic. Forms are embeddable, collect submissions, and integrate with webhooks.

The Problem

Form builders require manual configuration: dragging fields, setting validation rules, wiring up conditional logic. For non-technical users, this is still complex. For developers, it's tedious. Both groups want to describe what they need and get a working result. Existing AI tools generate static HTML forms but miss validation, file uploads, conditional visibility, and submission handling.

Approach

Natural language to form schema

The user types a description. An LLM parses it into a structured JSON schema: field types, labels, validation rules, conditional visibility, and layout hints. The schema is deterministic (same description produces same output) thanks to structured output parsing with strict schemas. Users can edit the generated schema through a visual editor after generation.

Runtime form renderer

Forms render from the JSON schema at runtime. The renderer handles text inputs, dropdowns, file uploads, date pickers, rating scales, and multi-step layouts. Conditional logic evaluates client-side: field B only shows if field A equals a specific value. Validation runs both client-side (instant feedback) and server-side (tamper-proof).

Submissions pipeline with integrations

Every submission is stored, validated, and optionally forwarded via webhook to external tools (Notion, Google Sheets, Slack, email). File uploads go to S3-compatible storage with presigned URLs. The submissions dashboard supports filtering, CSV export, and response analytics.

Challenges

LLM output reliability for form schemas

LLMs occasionally generate invalid schemas: missing required fields, impossible validation rules, or circular conditional logic. Built a validation and repair layer that catches structural errors, auto-fixes common issues (like a required field with no label), and rejects schemas that can't be repaired. The repair layer catches 98% of issues without re-prompting.

Embed performance on third-party sites

Embedded forms must load fast and not interfere with the host page. Built the embed as an iframe with a 12KB loader script. The form renders inside the iframe with its own isolated styles. Communication with the host page uses postMessage for submission events and height auto-adjustment.

Results

Formwise has generated over 12,000 forms since launch, with an average generation time of 3 seconds from description to working form.

12K+

Forms generated since launch

3s

Average generation time

98%

Schema validation pass rate

Tech Stack

Next.jsForm builder UI, submissions dashboard, and embed hosting
TypeScriptType-safe form schema definition and validation engine
PostgreSQLForm definitions, submissions, and user accounts
OpenAINatural language parsing to structured form schemas
Cloudflare R2File upload storage with presigned URLs for submissions
Tailwind CSSForm renderer styles and dashboard UI