Dublin, Ireland
Udemy Inc.
Dublin City UniversityOverview
Hi, I'm Ciarán 👋
Thanks for taking the time to check out my portfolio - I'm excited to connect with like-minded people.
Background
I've always been intrigued by language and technology. As a child, I spent my time learning languages and discovering how systems work. My first computer was a linux laptop, this is where my journey into programming began.
In my first graduate job as a Localization Coordinator at Udemy; I started learning how to code using our platform. Realizing how powerful automations and integrations are, I started programming workflow automations and integrating platforms via APIs in App Scripts to save hours of manual work per week.
Now, I design and build AI integration pipelines into our tech stack for internal, custom solutions using next-generation technology.
Core Devotions
- Linguistics
- Machine Learning & AI
- Formula 1
- Learning
- Training
Languages
- EnglishNative
- JapaneseN2 JLPT
- FrenchC1
- DutchB2
- IrishB1
Experience
Professional & Academic Background
Work
Machine Learning Engineer
Udemy
Hybrid software engineer and project manager. Designed, built and deployed production MT platforms, APIs and services with machine learning models.
Program Manager
Workday
Worked on implementing project management tools for Workday's globalization team and begain projects around connecting legacy data lakes to clear data and tracking.
Education
BA Applied Language & Translation
Dublin City University
Advanced Japanese and French with honors. Actively involved alumni initiatives including guest lecturing on localization in tech.
International Exchange
大阪公立大学 | Osaka Metropolitan University
Specializing in translation technology, CAT tooling, and scientific/economic localization.
Technology & Frameworks
Software engineering tools and frameworks I've the most experience in
Python
Kotlin
JavaScript
Rust
PostgreSQL
Redis
MongoDB
Firebase
PyTorch
TensorFlow
Modal
OpenAI Platform
Personal Projects
Some of my personal projects based around my interests in AI, automation and machine learning
Pumatic Neural Machine Translation
NMT Interface
Pumatic NMT a modern translation interface for public usage of my neural machine translation models. Translations are currently available from English to French, German, Spanish and Polish. Both text-to-speech and speech-to-text functions are available.
Custom Models and Inference
I trained models both from scratch and fine-tuned existing models based on Helsinki's open source MarianMT models. Used Modal's serverless API for model training with concurrent H100 and H200 GPUs on open source datasets. Training technologies include PyTorch, MarianMT and Tensorflow among others.
Fully Loaded Platform
Full seperate inference and backend services running behind the scenes with OAuth via Google and GitHub, translation history tracking, API key generation, and API & UI usage tracking in the user dashboard.
Core Technologies
Native AI File Sorting Agent
Use Case
A program for sorting files based on their content with native AI integration by using Apple's Vision framework.
How it Works
Launched in terminal, the user can add customizations to a config file to define which directories folders will be watched, the database of the folders, ignore filetypes. The program then matches the file type and places it in the database directory within a folder of their file domain such as 'IMAGE' or 'VIDEO'
Local AI Integration
Currently native to Apple Intelegence, for photos and other files, it uses Apple's Vision framework via objective-c foundations to rename images to be relitave to its content. More filetypes and AI configurations such as Ollama for Linux and Windows users are in the works.
Core Technologies
High-Performance Inference Service
Custom Architecture
A custom API modeled my own NMT models. Currently available from English to French, German, Spanish and Polish
Public API
Available for public use at https://pumatic.eu/docs. API keys can be generated in the user settings. API limit tracking can be seen both via API and the user dashboard. Limits are in place for proof of concept, but I am always happy to increase these on request.
Custom Inference & Backend
Custom inference and seperate FastAPI-based backend for cost-effective deployment and performance.
Core Technologies
Sync Google Calendar Events to Notion Without a Subscription
Completely Free
With full instructions and code available on GitHub, it gives the user a free solution for world problem. Deployed in Google App Scripts for the ultimite custom integration.
Solution
Adds your Google Calendar events to a Notion database. Something Notion themselves have not done yet. Third parties like Zapier charge users a subscription for this exact solution where this is free, open source and easy to implement.
Smart Features
Smart features pre-implemented such as Zoom link extraction, HTML parsing for metadata, avoids duplicates and processes events in advance.
Core Technologies
Professional Impact Projects
With a focus on highly scalable, internal solutions without the vendor cost and overhead. I specialise in creating custom solutions by building creative, next generation technology from scratch
AI-Powered TMS Machine Translation System
Completely internal system for machine translation of Translation Management System (TMS) content and user-content, replacing paid API services like DeepL. Includes complex XLIFF parsing, prompt engines, prebuild metadata in async pipelines via the TMS API. Complex RAG systems via vector databases. This internal system saves over 200K USD annually.
Machine Learning Quality Assessments
Using agentic AI models via API, I've designed and built programs for Linguistic Quality Assessment (LQA) where LLMs are used to generate quality reports on linguistic and functional quality in real production environments. Running completely in the background, it allows localization PMs to focus on production tasks and saves on expensive LQA costs. Issues are then triaged into engineering teams and linguists to review and resolve.
OpenAI Model Fine-Tuning
Using OpenAI's Enterprise Platform, I've fine-tuned existing GPT models with our legacy metadata such as translation memories and term bases, indexed into JSONL format for native training. This creats specific models we can use in other programs such as our TMS machine translation system for ultimate quality, customization and cost savings.
Localization Automation Suite
Built a comprehensive stack of automations to handle internal localization workflows to reduce manual work both internally and on the vendor side; saving workload capacity and LSP costs. Some of these include: Vendor Notification System for CL Projects, Automated TMS Workflow from Upload to Completion, Middleware Platforms for Live and Dynamic Translation Project Statuses to Stakeholders in the Request Tickets, and Automated Project Delivery via API to PM Systems.
Live Translation
Try the live translation demo below, powered by my NMT models running on the same server as Pumatic
Get in Touch
Have a project in mind or just wanna network? Send me a message and I'll definately get back to you
Connect With Me
Find me on these platforms
