Software

Two platforms I've built by leveraging AI.

Here is how I developed each one, from a real problem to a working product: Custom Creative, the company I founded, and Orac Voice, a free app for private dictation.

  1. 01Custom CreativeFounder
  2. 02Orac VoiceFree, open-source app

01Founder2026customcreative.pro

Custom Creative

Your car. Any scene. No shoot.

What it is

AI-generated advertising photography for car brands. The brand's vehicle is set up once, with its proportions, paint, wheels and badges locked.

From then on, every photo is directed like a shoot: factory color, camera angle and height, light, a 24, 50, 70 or 90 mm lens, a scene described in plain words and a retouch asked for in one sentence. Each image is delivered in 2K or 6K, in native 16:9, 21:9, 9:16, 4:5 and 1:1, always with the whole car in frame.

What I did

I founded the company in 2026 and directed:

  • The product, and the prompt system that keeps every vehicle accurate.
  • The studio interface.
  • customcreative.pro, built around a dial that turns the camera around the car.
  • How it works, a 30-second film in English and Spanish, 16:9 and 9:16, made with HyperFrames.
Visit customcreative.pro
The dial

One car. Nine angles. Nine scenes.

One generated Mercedes-Benz GLE 350 at nine verified camera angles, each generated in another scene and color.

  1. Black Mercedes-Benz GLE 350, seen from the front. Scene: Viña del Mar seafront.
    0°BlackViña del Mar seafront
  2. White Mercedes-Benz GLE 350, front three-quarter view. Scene: Kennedy highway.
    22.5°WhiteKennedy highway
  3. White Mercedes-Benz GLE 350, front three-quarter view. Scene: Apoquindo Avenue.
    45°WhiteApoquindo Avenue
  4. Silver Mercedes-Benz GLE 350, right side profile. Scene: Old-town street.
    90°SilverOld-town street
  5. Silver Mercedes-Benz GLE 350, rear three-quarter view. Scene: Valparaíso hills, night.
    135°SilverValparaíso hills, night
  6. Dark blue Mercedes-Benz GLE 350, seen from behind. Scene: Santiago riverside.
    157.5°Dark blueSantiago riverside
  7. Silver Mercedes-Benz GLE 350, left side profile. Scene: Vitacura, night.
    270°SilverVitacura, night
  8. White Mercedes-Benz GLE 350, front three-quarter view from the left. Scene: Atacama desert.
    315°WhiteAtacama desert
  9. Black Mercedes-Benz GLE 350, front view, slightly from the left. Scene: Wet asphalt.
    337.5°BlackWet asphalt
Retouch

One sentence. New paint. Same shot.

White Mercedes-Benz GLE 350 on the Kennedy highway, before the retouch.

Before White

InstructionChange the paint to black.

The same shot after the retouch: the GLE 350 is now black.

After Black

Film

How it works · 30 s

English and Spanish, 16:9 and 9:16. Made with HyperFrames.

Every vehicle image here was generated with Custom Creative. Brands and models belong to their manufacturers.

02Free, open-source appmacOS + Windows2026GitHub

Orac Voice

Say it. Get clean text. Keep it private.

01 / The Problem

Private dictation was not optional.

My wife needed speech-to-text for work, but the sensitive information she handles ruled out subscription-based tools that send data to external servers. She needed something useful that could stay on her own computer.

02 / The Solution

A local pipeline, built around the way she works.

Using Claude Code, I paired whisper.cpp for on-device transcription with a lightweight Ollama model that restores punctuation, removes filler words and turns raw dictation into clean, usable text. I then designed a simple interface that anyone can configure in seconds.

03 / The Result

Speak naturally. Get polished text. Keep control.

Orac Voice is now a free local-first app for macOS and Windows. It can be installed with Claude Code, Codex or directly from the terminal, and the default workflow keeps voice data on the user's machine.

Product walkthrough

See how Orac Voice works.

Watch directly on Loom

The finished experience is deliberately simple: hold a key, speak and release. Clean text appears wherever the cursor is, ready to send, edit or keep working with.

  1. Hold a key and speak
  2. Whisper transcribes locally
  3. Ollama cleans the text
  4. Clean copy appears at your cursor
Download free on GitHub
Get in touchLet's work together.