I’ve been a solo founder for over a year, building an AI-based service. Now, I’m looking to become a team player and learn from working with others.

Skills/Tech

Field
Technologies
Proof
Sales & Customer Support
Cold email, call and text outreach, Customer support
About 15 chess tournaments sold and managed in Spain, France, and Germany (most through cold outreach) this summer (2024): moveread.com, tnmt.moveread.com,
Deep Learning
Pytorch, TF. Background: probability, linear algebra, and calculus
github.com/marciclabas/tensorflow-ocr, github.com/marciclabas/mltools, https://github.com/moveread/ocrmap
Computer Vision
OpenCV. Background: computational geometry
github.com/marciclabas/cvjs, github.com/marciclabas/react-image-manipulation, github.com/marciclabas/pure-cv
Backend
Python, FastAPI
api2.moveread.com/docs
Webdev
React, Typescript, NextJS
github.com/marciclabas/react-chakra, marcelc.uk/, github.com/marciclabas/actually-beautiful-dnd, github.com/marciclabas/contextize
Sysadmin
Bash scripting, Cron, Services, etc.
Hard to prove, but: github.com/marciclabas/cit
Devops
Docker, Azure
https://hub.docker.com/u/moveread, https://github.com/marciclabas/func-az. Most sites mentioned above are hosted on Azure SWA, backends are Docker on Azure Web Services. Data of my service is stored on Cosmos DB and Azure Blob (https://github.com/moveread/core)
Other stuff I know but haven’t used in production: C, Rust, Haskell, Android (Kotlin w/Jetpack), Java

Formal Education

BS in Informatics Engineering

  • Faculty of Informatics of Barcelona (FIB) of the Polytechnical University of Catalonia (UPC)
  • Average: 9.01/10 overall and 9.7/10 in the initial phase
  • Currently in last (4th) year

Professional Music Degree

  • Municipal Music Conservatory of Barcelona (CMMB)

French DELF B2

Projects

Moveread

A service for chess tournaments. This summer (2024), it has been used in over 15 tournaments across Spain, France and Germany.
Moveread
An AI-based app to digitize handwritten chess scoresheets
Moveread
https://moveread.com/en/
Moveread involved training these models:
Usage
Model
Details
Data
Image Preprocessing
Table Transformer
Adapted to predict contours (instead of bounding boxes)
Finetuned on over 6k annotated scoresheet images
OCR
CNN+RNN (with CTC)
ㅤ
Finetuned on over 500k annotated handwritten chess moves
OCR Postprocessing
BERT
(Position, Notation) → Move. Trained with related tasks and simulated errors and OCR noise
Trained from scratch with increasing difficulty. Millions of positions, randomly generated and from real online games

Open Source

Many open source libraries as a by-product of Moveread:
Package
Language
Description
haskellian
Python
A method-chaining style Functional Programming library
kv
Python
An async dict interface, with implementations for SQLite, Postgres, Redis, Azure Blob, Azure Cosmos DB, the file system, and more
react-chakra
React TS
Animations and user interaction components over ChakraUI and Framer Motion. Live demo
react-image-manipulation
React TS
Perspective cropper and grid selector UIs, over FabricJS and OpenCVjs (via web workers). Live demo
And more! GitHub