Bogdan Ouatu
Software Engineer
Bogdan Ouatu
Bucharest, Romania
Email: bogdan@ouatu.ro
Website: ouatu.ro
GitHub: github.com/ouatu-ro
ㅡ
Skills
Product Strategy and Development • Leadership and Management of Small Teams • Cloud-Native Architecture Design • Technical Innovation and Problem-Solving • Client and Stakeholder Collaboration • Machine Learning Solutions Development • Financial Solutions Development
ㅡ
Experience
CTO & Co-Founder / Verbalate LTD
Jul 2023 - PRESENT
Led the overall product strategy and development for verbalate.ai, a cutting-edge video translation platform, scaling its user base from inception to mid-ten thousands.
Managed and nurtured a small team, hiring and contracting developers, while closely aligning the team's efforts with the company's long-term vision and client needs.
Designed and implemented a cloud-native, event-driven architecture to ensure scalability and flexibility, enabling the platform to support thousands of daily active users with high reliability and performance.
Oversaw the development of machine learning models, including a lip-syncing model that delivered superior quality and accuracy for unconstrained videos, setting a new benchmark at the time of deployment.
Pioneered a multilingual speaker diarization pipeline, setting a new standard for accuracy in unconstrained audiovisual media.
Enabled seamless user experiences by creating an in-browser translation editor, allowing users to proofread and edit their translations in real-time.
Collaborated with stakeholders and clients to continuously refine the product, ensuring it meets market demands and drives business growth.
Consultant / Freelance developer
Jan 2020 - Jul 2023
As interim CTO, guided the development of a speech processing system. My contributions directly led to deploying a functional system in three weeks for a team stuck in research and paper implementation for 6 months
Designed and developed an analytics and back testing app for trading stocks and options, decisive for choosing the main investment strategy of the client.
Developed Forex scripts and indicators for retail investors to help them generate consistent alpha for two months.
Created business intelligence (BI) apps in Vue and Streamlit to explore natural language data, such as customer support logs and net promoter score forms, leading to substantial reprioritization of the client’s business.
Built a location-based money transfer platform in Django as a proof of concept for a finance startup to showcase their Open Banking APIs; the proof of concept was selected for the Central Europe Open Banking competition and won a special prize from Google.
Adore Me / Artificial Intelligence Engineer
Jun 2021 - Aug 2022
Developed and deployed the company's first data-driven recommender system that is currently in production and makes around 15–20% of recommendations for operations worth around $120 million per year in revenue.
Created an aspect-based sentiment analysis algorithm to gather product preferences directly from clients' wishes and automatically integrate them into the recommender system.
Built a BI app on top of the aspect-based sentiment analysis algorithm to help product owners prioritize work by letting them mine and analyze submitted forms from clients, substantially changing the focus of two teams.
Developed a similarity and clustering color algorithm to find complementary products and comparable items when predicting the performance of to-be-launched garments.
Built a Google Calendar data analytics pipe to compute and monitor time spent in team meetings to identify harmful patterns and potential burnout for a few colleagues.
The University of Bucharest / Graduate Teaching Assistant
Feb 2021 - Aug 2022
Taught Formal Languages & Automata Theory to undergraduate students.
Organized and held laboratories where students learned to implement different domain specific algorithms.
Used GitHub classroom to track and manage assignments and automate assignment grading.
Adore Me / Full-stack Developer
2020 - 2021
Built an influencer marketing platform to allow the company to spend an extra $9 million per year on marketing without increasing the customer acquisition cost.
Developed a data scraping app to gather prospects to integrate and automate Instagram messaging in the app before any API support, which helped the company reach 100,000 prospects in one year.
Handled feature development, database modifications, bug fixing, code reviews, deployment, CI/CD with GitHub Actions, and server management while creating and solving tickets on Notion and Jira.
ㅡ
Projects
Parse Bank Statements
https://github.com/ouatu-ro/parse-bank-statement
Many Romanian banks currently do not export data in a friendly format, so parsing PDF bank statements is required. The project was about a simple script parsing bank statements for Raiffeisen and Alpha Bank. Every financial department can use it for easier auditing transactions made on company credit cards, saving 25 hours a year of error-prone work. I also use it to track my expenses.
Python Dependency Scripts
https://github.com/ouatu-ro/python-dep-scripts
This project provides terminal commands to facilitate some of the most challenging problems in the Python ecosystem, including Python version management, virtual environments, and dependencies. It ties together two other tools, pyenv, and Poetry. For more details, the README file in the GitHub repository reads as a tutorial and an article.
Covid Data Visualization
https://ouatu-ro-covid-data-viz-covid-bky2ls.streamlit.app/
This project aims to provide data regarding vaccination, variants, and notification rate to be easily seen and understood. The demo app offers many options, such as comparing data from multiple countries, selecting preferred plots for data visualization, and displaying the official PDFs from the European Centre for Disease Prevention and Control web page containing a data description.
Eletron Voice Cloning
A state-of-the-art (with regard to how natural the voice sounds) text-to-speech synthesis application for the Romanian language that can clone women's voices with less than one hour of speech. The cloned voice achieves a 4.47 +-.15 out of 5 naturalness score as measured by a medium human opinion score on the Likert scale compared to real voices' 4.80+-.14.
A direct comparison was made to the previous SotA and Google Translate Romanian voice. Over 90% of subjects find Eletron to have a more natural voice, and 100% of subjects find it more natural than Google Translate voice. Eletron is based on the Tacotron model and trained using GCP resources and the SWARA dataset.
ㅡ
Education
The University of Bucharest / Master's Degree in Artificial Intelligence
2020 - 2022, Bucharest, Romania
4.0 GPA
The Alexandru Ioan Cuza University / Bachelor's Degree in Computer Science
2016 - 2019, Iasi, Romania
4.0 GPA