Gilbert Pochet
Melbourne,Australia
Summary
Innovative AI leader with over 10 years of experience driving engineering teams across Australia, Brazil, and Germany. Delivered Generative AI and ML systems that enhanced publishing efficiency and cross-sell conversions. Designed and implemented AI/ML platforms using Generative AI, LLMs, and multimodal models, ensuring automation, personalization, and scalable cloud-native deployments on GCP and AWS.
Overview
13
13
years of professional experience
Work History
Staff Machine Learning Engineer
REA Group
, VIC
10.2025 - Current
- Designed NLP-driven search using embeddings and LLMs to enable semantic retrieval from unstructured text, improving relevance.
- Built multimodal search (image + text) combining visual and textual signals for more accurate query matching.
- Led development of CNN-based pipelines for large-scale image feature extraction, powering downstream search and ranking systems.
- Architected and deployed scalable ML pipelines on AWS (Lambda, SQS) using Kafka-based streaming for reliable data processing.
- Defined technical strategy, mentored engineers, and established AI-assisted coding standards (Cline/Copilot) to enhance code quality and collaboration.
Principal Machine Learning | CTO
JubJub
Australia
04.2024 - 09.2025
- Led the technology organisation, scaling a cross-functional ML and engineering team across backend, frontend, and infrastructure with zero attrition over 18 months.
- Engineered a frame-based multimodal LLM pipeline with LangChain to auto-generate video captions and social copy, reducing publishing time by 21%.
- Experimented with LoRA and ControlNet on SDXL/LLaVA/PixArt, deploying via vLLM to achieve high-throughput, KV-efficient multimodal generation.
- Architected ML platform on GCP with modular pipelines (Cloud Tasks, Docker, Kubernetes) to streamline ingestion, transformation, evaluation, and publishing of multimodal content from YouTube, TikTok, LinkedIn, and other APIs.
- Built Guardrails and LangChain Eval pipelines to validate LLM outputs, enforce safety protocols, and track quality using Langfuse.
- Defined and enforced MLOps practices: reproducible environments with Poetry, MLflow-style experiment tracking, scalable deployment via FastAPI and Kubernetes.
- Architected and integrated endpoints from Meta, YouTube, TikTok, LinkedIn, and Vimeo into secure, scalable ingestion and publishing pipelines.
- Founded a startup focused on innovative technology solutions.
Staff Machine Learning Engineer (Contract)
oOh!media
Melbourne, Australia
02.2023 - 05.2024
- Developed deep learning model for real-time campaign reach classification, enhancing accuracy and consistency compared to manual heuristics.
- Designed and maintained scalable data architecture on Databricks and PySpark to power real-time audience segmentation and programmatic advertising systems.
- Engineered full ETL pipelines for campaign data ingestion, cleaning, and feature generation, ensuring reliable model training and inference.
- Architected event-driven pipelines integrating external datasets with internal systems, ensuring data quality, lineage, and real-time processing of audience signals for predictive modelling.
- Standardized CI/CD and MLOps practices with MLflow and Databricks, improving reproducibility, deployment speed, and observability.
- Implemented scalable hyperparameter tuning with Optuna, accelerating experimentation and model optimization.
- Mentored engineers and data scientists on scalable system design and production ML deployment, strengthening team capability and technical excellence.
Principal Data Scientist
Mr Yum
Australia
01.2022 - 02.2023
- Architected personalization and upsell systems, integrating LLM-based classification and hybrid recommendation pipelines that drove a 24.9% increase in total order value.
- Developed hybrid recommender system integrating item-item collaborative filtering, TF-IDF weighting, and TinyBERT classifier for menu item categorization, improving relevance and complementarity of suggestions.
- Designed and maintained nightly retraining pipelines in PySpark on Databricks, processing 250k+ daily transactions with UDF-based TinyBERT inference and dynamic menu updates.
- Established scalable MLOps practices using Databricks, MLflow, and Poetry-driven modular services, improving reproducibility, deployment speed, and traceability of experiments across teams.
- Defined the strategic AI/ML architecture roadmap, balancing flexibility, security, and scalability, while enabling experimentation and aligning quarterly reviews with evolving product needs.
- Enhanced data platform with Hugging Face LLMs, enriching product catalogs and improving user engagement.
- Designed retraining pipeline on Databricks using PySpark + MLflow, processing 250k+ daily transactions.
- Startup
Principal Data Scientist
SEEK
Australia
07.2018 - 01.2022
- Pioneered use of transformer-based LLMs (XLNet, Longformers) to generate contextual embeddings for resumes and job descriptions.
- Designed composite embeddings blending semantic vectors with structured features (tenure, seniority, skill density) for accurate, business-relevant matching.
- Built a semantic recommendation system that boosted recruiter contacts by 10.3% while reducing bias through fairness-aware preprocessing.
- Scaled retrieval with dimensionality reduction (Random Projection) and Elasticsearch to enable sub-second search across millions of candidates.
- Developed monitoring and evaluation frameworks for drift detection, recruiter engagement metrics, and offline benchmarking to enhance decision-making.
- Mentored junior scientists to adopt deep learning and NLP/LLM techniques, fostering skill growth and innovation within teams.
Senior Data Scientist
X24Factory
Berlin, Germany
12.2016 - 07.2018
- Led strategy and execution of AI projects in data science department, aligning initiatives with business goals.
- Developed NLP and deep learning solutions with TensorFlow, Gensim, and SpaCy, enhancing capabilities in CNN, NER, and word embeddings.
Senior Data Scientist
MICE Portal GmbH
Berlin, Germany
08.2015 - 09.2016
- Applied machine learning techniques to develop prototypes for sales and marketing optimization.
- Built Random Forest models and NLP pipelines to analyze product and sales data for actionable insights.
Data Scientist
Stayfilm
Sao Paulo, Brazil
06.2013 - 08.2015
- Developed topic modeling and hierarchical classification tools in Java for content categorization.
- Extended NLP capabilities to support English, Portuguese, and Spanish for improved clustering.
Education
Master - Information Engineering
UFABC - Federal University Sao Paulo(ABC Region)
São Paulo, Brazil08-2014
Bachelor - Computer Science
Universidade Internacional de las Americas
San Jose, Costa Rica06-2012
Skills
- Leadership
- VLLM
- SGLang
- KV Cache
- LangChain/Pydantic
- AI
- RAG
- Vector Search
- MCP
- Machine Learning
- NLP
- Multimodal/Generative Models
- LangFuse/MLflow
- GCP
- AWS
- Kubernetes
- Poetry/UV
- Python
Websites
Publications
- Analysis of influence through a social Network (Facebook) to determine whether ideas of friends to friend can be distributed along the network.
- Presented at IBM Research Colloquium 2013 with the Poster 'Can Social Network Influence Voters.'
- Revealing User Influence inside Online Newspaper.
Personal Information
Nationality: Australian Citizen
Timeline
Staff Machine Learning Engineer
REA Group
10.2025 - Current
Principal Machine Learning | CTO
JubJub
04.2024 - 09.2025
Staff Machine Learning Engineer (Contract)
oOh!media
02.2023 - 05.2024
Principal Data Scientist
Mr Yum
01.2022 - 02.2023
Principal Data Scientist
SEEK
07.2018 - 01.2022
Senior Data Scientist
X24Factory
12.2016 - 07.2018
Senior Data Scientist
MICE Portal GmbH
08.2015 - 09.2016
Data Scientist
Stayfilm
06.2013 - 08.2015
Master - Information Engineering
UFABC - Federal University Sao Paulo(ABC Region)
Bachelor - Computer Science
Universidade Internacional de las Americas