Jan 2023 – Dec 2023 · 11 mos · Berlin, Germany · Remote

• Sequoia Open Source Fellowship: Keeping an Open-Source Mind: https://www.sequoiacap.com/article/sebastian-ramirez-spotlight/
• Sequoia Open Source Fellowship: https://www.sequoiacap.com/oss/

Jun 2022 – Dec 2022 · 6 mos · Berlin, Germany

• 🤓🚀 https://www.forethought.ai/press/forethought-hires-creator-of-fastapi-sebastian-ramirez

May 2021 – Jun 2022 · 1 yr 1 mo · Berlin, Germany

• 🤓🚀 https://www.forethought.ai/press/forethought-hires-creator-of-fastapi-sebastian-ramirez

Jul 2019 – Apr 2021 · 1 yr 9 mos · Berlin Area, Germany

• https://explosion.ai
• Worked on Prodigy Teams, the cloud and privacy-centric version of Prodigy, the data annotation tool for Machine Learning datasets using active learning, with deep integrations with spaCy, the open source Natural Language Processing toolkit.
• Also worked on many diverse projects, including the mypy plug-in for Thinc, the Deep Learning framework used by spaCy, internal APIs, and automation tools.

Jun 2018 – May 2019 · 11 mos · Dubai, United Arab Emirates

• Developed components of an interactive chatbot application, using a backend in Flask, a CouchDB database, and remote workers using Deep Learning based NLP techniques.
• Developed a system to store nationwide product prices data in Flask with a Couchbase database and a Vue.js frontend. Then migrated the backend from Flask to FastAPI.
• Developed an e-commerce system backed by a FastAPI backend connected to Couchbase and a Vue.js frontend.
• Several other projects that due to their nature remain classified. Mostly related to the application of Deep Learning systems for Computer Vision and NLP.

Mar 2018 – Jun 2018 · 3 mos · Dubai, United Arab Emirates

• Developed a multi-object detection system (using YOLO v3 re-implemented in Keras/TensorFlow), re-trained with custom business case categories using OpenImages data, converting annotations between COCO and Pascal VOC.
• Developed a system to record live images from a mobile application (written in Ionic/Angular) and location (GPS) data, and synchronize it with a backend server written in Flask with a CouchDB database, connected to a frontend visualization system in Angular.

Nov 2017 – Mar 2018 · 4 mos

• Developed advanced data processing tools and systems for classified projects in the government and telco sectors.

Apr 2016 – Oct 2017 · 1 yr 6 mos

• Directed the development of a Django application connected to PostgreSQL and Firebase, and then led the migration to Flask and CouchDB. Integrating it with a frontend application in Angular to handle human-realtime chat communications with semi-automatic Machine Learning interactions in the backend (with “human in the loop” techniques).
• Designed and implemented asynchronous architectures using RabbitMQ and Celery for background task processing, connecting workers with Tika (for OCR), and Deep Learning (for object detection/recognition) in a distributed system (leading a team of developers).
• Designed and implemented an automatic time tracking system for people working in computers, with a backend in Flask with PostgreSQL and Celery background jobs managed with RabbitMQ. Then migrated the database to ElasticSearch to handle the updated requirements and user base.
• Developed an Electron application in TypeScript with a local Redux state, that tracked time on each application run in the computer automatically and sent it to the backend system. The application was cross-platform (Windows, macOS, and Linux) and had automatic updates (with a custom updates server).
• Because of the nature of Senseta’s high tech products and clients, there are many other projects that remain “classified”.

May 2015 – Apr 2016 · 11 mos

• Migrated several “snowflake” servers (set up by hand) to more modern technologies: first automatized deployments with Ansible, then Docker, Rancher (Docker Compose), Docker Swarm mode. And migrated the stacks to these technologies, as well as introducing testing, CI/CD, and cluster monitoring.
• Developed a distributed video-wall screening system (several large TV screens working as a very large one) based on a central server and one Raspberry Pi for each screen.
• Developed a system to analyze telecommunications records in Spark, saving results to PostgreSQL that were then served by a Flask backend. Coordinated the frontend team that developed the user interface to show the data from that system.
• Developed a system to record audio from TV and detect the content with audio finger-printing, connected to a content search engine. Coordinated the development of the mobile application that captured the audio fragments.

Apr 2013 – Apr 2015 · 2 yrs

• Developed a corporate search engine based on Solr, created a desktop client application that scanned the user’s computer for documents and sent them to the server. These documents were then processed with Tika for data extraction, like contents of Word, Excel, PowerPoint, and PDF; and metadata in images like date, meta-tags, and location (for this project, no OCR was applied).
• Developed a sentiment analysis system using Twitter data, and a visualization system with maps based on topic (cluster), size, sentiment, and location of real-time data.
• Developed a computer vision system that analyzed raw video feeds from CCTV systems, detecting, tracking and counting people, even after occlusions. This was done using old computer vision techniques (deep learning was not widely available yet), including Lucas Kanade optical flow, SIFT and SURF keypoint extraction and matching, linear regressions for future position estimates live-trained on the last positions (because in tests it outperformed Kalman filters). The extracted image coordinates were then converted to the coordinates of the location plan (map) using a multivariate polynomial regression custom-trained for each camera/location. This technique was used over transformation matrices to account for changes in the space itself, like staircases. This system (and mainly the coordinate transformation component with Machine Learning) earned Senseta the presidential award “Innova”, given yearly to companies pushing forward technological advancements.
• Set up several Hadoop clusters with Hortonworks and Cloudera, automatized the installation and configuration across several machines, first with scripts, then with Ansible.