requinator

An AI-powered interview tool that conducts a structured conversation with potential consulting clients and automatically generates a Product Requirements Document from the responses.

Instead of a static briefing form, requinator runs a real conversation: the AI asks targeted questions one by one about problem, target audience, goals, scope, and context — up to 20, as few as necessary. At the end, it summarises everything into a structured PRD that the consultant can use immediately.

The PRD serves primarily as a briefing for the consultant and secondarily as a confirmation for the client that their requirements have been correctly understood.

Features

• Adaptive dialogue — targeted questions across six categories, adapting to each response
• Reference URLs — clients can attach URLs as inspiration; the backend fetches, extracts, and passes content as AI context
• Real-time streaming — AI responses appear token by token in the browser via Server-Sent Events
• Share link — every session has a persistent link; closing and reopening the browser fully restores state
• PRD download — finished document as Markdown, including a reference section with AI assessment
• Admin panel — protected overview of all sessions with transcript, references, and editable PRD

Architecture

Frontend without a build step — HTMX handles all dynamic interactions (streaming, refs list, PRD display) directly in HTML. No JavaScript framework, no bundler.

Lean backend — Express with clearly separated layers: routes for HTTP handling, services for business logic (LLM, interviewer, generator, refs, SSRF, rate limiting). SQLite as the single database file — backup is a single cp command.

AI integration via OpenRouter — model is swappable via environment variable. The interviewer logic controls the conversation flow through structured JSON in the system prompt; the generator produces the final PRD in a separate call.

Security — Reference URLs are checked against private IP ranges before each fetch (including after redirects) to prevent SSRF. Rate limiting on session creation, responses, and ref fetches. User responses and reference content are not logged.

Tech Stack

Component	Technology
Runtime	Node.js 22 LTS
Server	Express
Database	SQLite via better-sqlite3
Frontend	HTMX + Vanilla JS
Styling	Pico.css
Templates	EJS
AI Provider	OpenRouter / Claude Sonnet
Streaming	Server-Sent Events
Deployment	PM2 + Nginx
CI/CD	GitLab Pipelines

Deployment

GitLab pipeline with two stages: tests on every push, automatic deployment on main. The server runs behind an Nginx reverse proxy with buffering disabled — required for SSE streaming. PM2 handles process management and restart after server reboot.

---

Proprietary — all rights reserved.