Felix

4 modules from the current build: Pinout extraction (resolves exact package, rejects wrong candidates), Footprint geometry (IPC-7351B land pattern with source evidence), Register Map (firmware-ready C headers with Pos/Msk pairs), and KiCad CAD Bundle (.kicad_sym + .kicad_mod audited and import-ready).

Screenshots from the live platform at datasheetvend.com , extraction pipeline, dashboard, and the UCM prototype. Early access, active development.

A few screenshots from the current build , still heavy WIP but the core pipeline is running.

What i shipped , DatasheetVend

I'm building a platform that turns hardware component datasheets into structured, machine-readable engineering data.

You upload a PDF. It extracts pinouts, footprints, register maps, timing diagrams, thermal limits, absolute maximum ratings, analog specs, power profiles, logic levels, passive BOM data, and per-pin caution flags all as structured JSON, with the source page and evidence attached to every field so you know exactly where each value came from.

The core pipeline is live. PDF upload, Stripe-gated extraction, Cloud Run processing, and a growing set of 14 extraction modules. The pinout, footprint, thermal, and register map modules have public API v1 routes. You can already extract a real datasheet and get back a JSON pack you can actually pipe into KiCad, Altium, or your own tooling.

A lot it's being built. The platform is in heavy active development module coverage, output contract hardening, human review workflows, confidence scoring, full audit trails. We're building this to be trustworthy by design, not just fast.

The part i'm most excited about -UCM :

We're prototyping a Unified Component Model. Instead of returning separate JSON files per module, UCM fuses all extraction lanes at the individual pin level. A single query on Pin 1 returns its layout, voltage limits, alternate functions, analog range, and caution warnings all resolved together. Think pin-level digital twin. One unified object that Altium, KiCad, firmware pipelines, and AI agents can all consume without correlating fragmented arrays on-the-fly. No commercial tool does this from raw PDFs today.

Where it ends:

A trusted engineering data layer that engineers, CAD tools, firmware pipelines, and AI agents can all consume without re-reading the same PDF manually ever again.

Early access is live at Datasheetvend.com

Would love to connect with anyone in embedded, EDA, or hardware who's felt this pain.