Galdra

Hey IH,

  I'm a solo developer based in Spain. Over the past few weeks I built Galdra (https://galdra.dev) — a REST API that turns any PDF into structured JSON.

  The idea: you define a JSON schema (the fields you want), upload a PDF, and get validated data back. One API call. No templates to configure, no training data, no document-specific setup.

  Why I built it:

  Every PDF extraction tool I tried fell into two camps: enterprise platforms that cost $500+/mo and need weeks of setup, or template-based tools that break when the document format changes. I wanted

  something in between — just tell it what you need and get it back.

  How it actually works:

  1. OCR extracts text from the PDF

  2. An LLM structures the text according to your schema

  3. The result gets validated against your schema types

  4. If there are errors, it auto-repairs and retries (up to 3 attempts)

  5. Returns clean JSON with a confidence score

  The tech: FastAPI, PostgreSQL, Redis, Stripe billing, OpenRouter for LLM calls. BFF auth pattern with HttpOnly cookies. Deployed on a single VPS.

  Pricing: Free (200 pages/mo), Scribe $29/mo (3K pages), Forge $79/mo (10K pages). Each tier uses a more capable AI model.

  Current status:

  - Product is live and working

  - $0 MRR, 0 customers

  - Just set up Twitter and Crunchbase today

  - Planning Show HN and Product Hunt launches

  I'd love feedback on:

  - The pricing — does the free tier feel generous enough to try?

  - The landing page (galdra.dev) — does the value prop come through?

  - What would make you actually try this over just writing your own script with an LLM API?

  Happy to share anything about the build — architecture, costs, mistakes, whatever.