Ricardo is available for hire
Hire RicardoRicardo G. Sousa
Verified Expert in Engineering
Artificial Intelligence (AI) Developer
Porto, Portugal
Toptal member since May 13, 2024
Bio
Ricardo is an experienced AI and product strategy expert and a major eCommerce platform lead. He has created advanced automations and AI personalization, recommendation, and search models, generating an annual revenue uplift of over $16 million. With 10+ years of experience, Ricardo excels at driving the development and successful launch of innovative AI platforms at scale, from early-stage startups to mature enterprises, translating data-driven insights into strategic competitive leverage.
Portfolio
Continuum Pulse
Agentic AI, Amazon Web Services (AWS), API Integration, Algorithms...
SeeOnMe
UNet, Diffusion-based AI Models, MongoDB, Redis, Sentry...
Visor
AWS Command Line Interface (CLI), Python, PyTorch, Docker, Whisper...
Experience
- Python - 15 years
- Strategy - 15 years
- Artificial Intelligence (AI) - 15 years
- Machine Learning - 15 years
- Computer Vision - 10 years
- Data Engineering - 8 years
- Vector Search - 8 years
- Semantic Search - 8 years
Preferred Environment
Python, Artificial Intelligence (AI), Machine Learning, Strategy, Natural Language Processing (NLP), Vector Search, Conversational AI, Retrieval-augmented Generation (RAG), Data Engineering
The most amazing...
...thing I've built is a multimodal AI agent platform that blends computer vision and NLP, enhancing online shopping and increasing customer satisfaction by 30%.
Work Experience
AI Engineer & Data Scientist/Founder
2024 - PRESENT
Continuum Pulse
- Spearheaded the development and deployment of scalable AI solutions, utilizing GenAI models and cloud infrastructure to enable 10% business growth for early-stage international companies.
- Developed and deployed an AI-powered transcript assessment service using LLMs, RAG, and Chain-of-Thought prompting, increasing analysis efficiency by 40% and improving insight accuracy by 25% for a key client.
- Developed and implemented an AI-powered portfolio management tool using LangGraph and Qdrant, automating the risk assessment of new projects and generating data-driven insights and proposal designs, leading to a 20% increase in project success rates.
Technologies: Agentic AI, Amazon Web Services (AWS), API Integration, Algorithms, Vector Search, Semantic Search, Machine Learning, Artificial Intelligence (AI), PyTorch, Python, Streamlit, Qdrant, LangGraph, Image Analysis, AI Assistants, Troubleshooting, PostgreSQL
Chief AI Officer
2024 - 2025
SeeOnMe
- Drove the creation of a key business offering by developing an open-source diffusion model from the ground up, generating a wide range of Virtual Try-On (VTON) images.
- Designed, built, and deployed an AI platform that handled high-volume requests by leveraging scalable and resilient machine learning services and integrating seamlessly with AWS cloud services.
- Increased AI platform reliability by 20% by integrating Sentry for real-time error monitoring, performance tracking, and system health insights, proactively reducing service disruptions.
Technologies: UNet, Diffusion-based AI Models, MongoDB, Redis, Sentry, AWS Command Line Interface (CLI), Amazon Web Services (AWS), Python, Git, Docker, FastAPI, Image Analysis, Troubleshooting
Principal AI Engineer
2023 - 2024
Visor
- Designed, built, and deployed a scalable AI platform using data augmentation and transformers like large language models (LLMs) and BERT, improving customer service email categorization by 40% and enabling rapid AI model integration and deployment.
- Headed a top-performing AI division, driving innovation in customer service call transcription and report generation with GenAI for a responsible AI audit solution, and ensuring timely product delivery.
- Mentored a remote team, resulting in a 75% reduction in development cycles.
- Leveraged XP/Lean startup methodologies and orchestrated high-impact initiatives by designing robust frameworks for testing, learning, and quality assurance to assess return on investment (ROI).
- Leveraged analytics and data-driven insights to shape product reformulations.
Technologies: AWS Command Line Interface (CLI), Python, PyTorch, Docker, Whisper, Generative Pre-trained Transformer 4 (GPT-4), OpenAI, DVC, Sphinx, C4 Model, PlantUML, LaTeX, Light LLMs, Large Language Model Operations (LLMOps), Large Language Models (LLMs), Data Augmentation, Natural Language Processing (NLP), Strategy, Communication, Lean Project Management, Remote Team Leadership, Git, Bash Script, Makefile, Transformer Models, FastAPI, Management, Team Leadership, Artificial Intelligence (AI), Machine Learning, Deep Learning, Bash, Jira, PyCharm, Data Engineering, Machine Learning Operations (MLOps), Docker Compose, Discord, Amazon Web Services (AWS), Scalable Web Services, Test-driven Development (TDD), APIs, OpenAI GPT-3 API, Generative Pre-trained Transformers (GPT), REST, API Integration, Hugging Face, Generative Artificial Intelligence (GenAI), Chatbots, Software Architecture, Data Science, Scikit-learn, OpenAI GPT-4 API, Fine-tuning, Embeddings from Language Models (ELMo), Semantic Search, Technical Leadership, Startups, MongoDB, Pandas, AI Agents, Predictive Modeling, AI Chatbots, Full-stack, Predictive Analytics, Prompt Engineering, Solution Architecture, AI Assistants, Troubleshooting, PostgreSQL
Principal Data Scientist
2021 - 2023
Farfetch
- Improved a PySpark ETL process on Databricks and a LightGBM re-ranker model to personalize the customer experience and product recommendations. Analyzed customer behaviors, such as lifetime value (LTV), for insight discovery.
- Headed a $2.5 million project to create multimodal AI agents, achieving a 30% uplift in customer positive responses.
- Built an ecosystem, resulting in new business knowledge, architecture for retrieval-augmented generation (RAG), and new AI models incorporating natural language processing (NLP) with LLMs.
- Designed and implemented A/B tests and experiments to evaluate personalization strategies' impact on key performance indicators (KPIs) like add-to-cart rates.
- Partnered with product management and senior leadership to provide actionable insights and strategic recommendations.
Technologies: Python, PyTorch, PySpark, Databricks, Docker, CircleCI, Sphinx, C4 Model, PlantUML, Elasticsearch, LaTeX, Google Cloud Platform (GCP), BigQuery, Data Augmentation, Vector Search, Information Retrieval, Artificial Intelligence (AI), Machine Learning, Natural Language Processing (NLP), Natural Language Generation (NLG), GitHub, GitHub Actions, Flask, Personalization, Recommendation Systems, Learning to Rank, Jira, Git, MLflow, Communication, Strategy, Remote Team Leadership, Lean Project Management, Linux, Complex Problem Solving, Creative Problem Solving, Large Language Model Operations (LLMOps), Large Language Models (LLMs), Makefile, Deep Learning, Team Leadership, Management, Conversational AI, Bash, Bash Script, Search, PyCharm, Data Engineering, Retrieval-augmented Generation (RAG), Open-source LLMs, Machine Learning Operations (MLOps), Docker Compose, Zoom, Discord, Scalable Web Services, Test-driven Development (TDD), APIs, OpenAI GPT-3 API, Generative Pre-trained Transformers (GPT), REST, API Integration, Hugging Face, Generative Artificial Intelligence (GenAI), Chatbots, Software Architecture, Data Science, Scikit-learn, OpenAI GPT-4 API, Fine-tuning, Embeddings from Language Models (ELMo), SQL, Semantic Search, eCommerce, Technical Leadership, XGBoost, LightGBM, Vector Databases, MongoDB, Pandas, AI Agents, Predictive Modeling, Agentic AI, AI Chatbots, Taxonomy, Full-stack, Predictive Analytics, Prompt Engineering, Solution Architecture, Image Analysis, AI Assistants, Troubleshooting, PostgreSQL
Data Science Manager
2019 - 2021
Farfetch
- Leveraged novel approaches to optimize user experience, including customized segmentation and faceted search enhancements. Proposed impactful A/B tests that delivered an outstanding $14 to $16 million increase in potential prospects.
- Secured $2.4 million in collaborative funding from the Portuguese government to pioneer the development of the inaugural multimodal conversational AI agent.
- Uncovered insights in clickstream data using LightGBM models to enhance relevance in normalized discounted cumulative gain (nDCG) while aligning with business goals through personalized recommendations and faceted search.
Technologies: Python, Information Retrieval, Machine Learning, Artificial Intelligence (AI), Elasticsearch, PT-Ranking, Docker, C4 Model, PlantUML, Google Cloud Platform (GCP), BigQuery, Jenkins, Octopus Deploy, Deep Learning, Clustering, Bash, Team Leadership, Management, Conversational AI, Complex Problem Solving, Creative Problem Solving, Strategy, Makefile, Personalization, Bash Script, Search, Lean Project Management, Jira, PyCharm, Data Engineering, Machine Learning Operations (MLOps), Docker Compose, Zoom, Discord, Scalable Web Services, Test-driven Development (TDD), APIs, REST, API Integration, Chatbots, Software Architecture, Data Science, Scikit-learn, Fine-tuning, SQL, Semantic Search, eCommerce, Technical Leadership, XGBoost, LightGBM, MongoDB, Pandas, Predictive Modeling, AI Chatbots, Taxonomy, Full-stack, Predictive Analytics, Solution Architecture, Image Analysis, Troubleshooting
Data Scientist
2015 - 2019
Farfetch
- Led two teams and managed eight data scientists, conducting 15 A/B tests on Farfetch's listing pages. This resulted in search enhancements that drove a projected annual revenue increase from $6.2 to $8.7 million, with a 60% overall success rate.
- Secured the first chatbot team. Introduced a new service leveraging NLP pipelines and techniques such as tokenization, stemming, lemmatization, and word embeddings for ticket categorization, i.e., XGBoost, reducing average reply time by 28%.
- Fostered university partnerships, improving the DYM methodology with a projected annual gross merchandise value (GMV) uplift of $0.4 million and automated product categorization.
- Headed A/B testing on listing pages, estimating $620 thousand to $6.2 million in annual revenue gains. Implemented AI/ML solutions using Python for N-gram-based and point-wise faceted searches.
- Disrupted the platform with a novel functionality to visually search in the product catalog, contributing significantly to customer engagement in the marketplace.
Technologies: Python, Elasticsearch, Docker, Artificial Intelligence (AI), Machine Learning, Learning to Rank, Jenkins, Octopus Deploy, Computer Vision, Information Retrieval, BigQuery, Vector Search, Makefile, Team Leadership, Management, Bash, Bash Script, Search, Personalization, Jira, PyCharm, Data Engineering, Machine Learning Operations (MLOps), Docker Compose, Scalable Web Services, API Integration, Chatbots, Software Architecture, Data Science, Scikit-learn, Fine-tuning, SQL, Semantic Search, eCommerce, Technical Leadership, Startups, XGBoost, LightGBM, MongoDB, Pandas, Predictive Modeling, AI Chatbots, Full-stack, Predictive Analytics, Solution Architecture, Image Analysis, Troubleshooting
Experience
Video Shorts Generator
https://github.com/rjgsousa/video-shorts-generatorThe Video Shorts Generator (VSG) is designed to help businesses enhance their online presence by streamlining video and blog content creation. The objective was to build a system from scratch within one week that could generate clips from long video recordings, enabling lead generation on brand marketplaces and improving productivity.
Integrating NLP, video analysis, and generative AI, VSG simplifies video creation while delivering relevant shorts. Users can swiftly create Instagram-ready clips while eliminating extensive video editing. Additionally, VSG generates blog articles based on extracted content.
My contributions include:
• Developing an end-to-end ecosystem and platform, resulting in new business knowledge and processes.
• Architecting RAG and GenAI components for seamless language model integration.
• Leveraging theme segmentation techniques for efficient, relevant clip generation.
• Implementing automations throughout the platform to streamline content creation and curation workflows.
The successful deployment of VSG has enabled the client to streamline content creation, increasing efficiency, cost savings, and audience engagement.
Integrating NLP, video analysis, and generative AI, VSG simplifies video creation while delivering relevant shorts. Users can swiftly create Instagram-ready clips while eliminating extensive video editing. Additionally, VSG generates blog articles based on extracted content.
My contributions include:
• Developing an end-to-end ecosystem and platform, resulting in new business knowledge and processes.
• Architecting RAG and GenAI components for seamless language model integration.
• Leveraging theme segmentation techniques for efficient, relevant clip generation.
• Implementing automations throughout the platform to streamline content creation and curation workflows.
The successful deployment of VSG has enabled the client to streamline content creation, increasing efficiency, cost savings, and audience engagement.
Face Liveness and Age Detector
https://github.com/rjgsousa/face-liveness-age-detectorThe Face Liveness and Age Detector (FLAD) is a computer vision system that enhances security and authentication by detecting and analyzing faces in images and videos. It can be incorporated into a security platform to determine the authenticity of detected faces, i.e., liveness detection, and estimate the age of individuals.
It addresses the growing need for robust facial recognition and authentication solutions, particularly in industries where identity verification is critical, such as financial services or governmental institutions.
My contributions were on:
• The architecture of the system's liveness detection module, leveraging cutting-edge deep learning techniques to accurately differentiate between genuine and spoofed faces.
• The optimization of the system's performance to ensure efficient and scalable deployment.
• The development of user-friendly APIs to seamlessly integrate the FLAD into existing systems and workflows.
The successful deployment of the FLAD can enable organizations to enhance their security. By accurately detecting spoofed faces and estimating age, the FLAD has proven invaluable in safeguarding sensitive information and ensuring the integrity of authentication processes.
It addresses the growing need for robust facial recognition and authentication solutions, particularly in industries where identity verification is critical, such as financial services or governmental institutions.
My contributions were on:
• The architecture of the system's liveness detection module, leveraging cutting-edge deep learning techniques to accurately differentiate between genuine and spoofed faces.
• The optimization of the system's performance to ensure efficient and scalable deployment.
• The development of user-friendly APIs to seamlessly integrate the FLAD into existing systems and workflows.
The successful deployment of the FLAD can enable organizations to enhance their security. By accurately detecting spoofed faces and estimating age, the FLAD has proven invaluable in safeguarding sensitive information and ensuring the integrity of authentication processes.
Multimodal Conversational AI Agents
Developed a multimodal conversational AI agent for the online high-end fashion marketplace Farfetch. By mimicking a fashion expert, it creates a seamless, human-like conversational experience and aims to:
• Scale Farfetch's business while maintaining customer satisfaction.
• Leverage conversational AI to improve conversion rates.
• Leverage data and knowledge to provide personalized fashion recommendations.
• Revolutionize online shopping in high fashion with a cutting-edge conversational agent.
I've led this $2.5 million project, developed the ecosystem and platform, and achieved a 30% uplift in positive customer responses. My contributions include establishing new business knowledge, architecture for RAG, novel AI models like dialog state tracking, natural language understanding (NLU) with large language models such as JointBERT, and synthetic dataset generation procedures.
The project has tackled challenges like tracking evolving shopping needs through conversation and relating product details to dialogues for better engagement. The successful development has enabled Farfetch to deliver a high-touch, personalized experience at scale, meeting modern consumers' expectations and reinforcing its market leadership.
• Scale Farfetch's business while maintaining customer satisfaction.
• Leverage conversational AI to improve conversion rates.
• Leverage data and knowledge to provide personalized fashion recommendations.
• Revolutionize online shopping in high fashion with a cutting-edge conversational agent.
I've led this $2.5 million project, developed the ecosystem and platform, and achieved a 30% uplift in positive customer responses. My contributions include establishing new business knowledge, architecture for RAG, novel AI models like dialog state tracking, natural language understanding (NLU) with large language models such as JointBERT, and synthetic dataset generation procedures.
The project has tackled challenges like tracking evolving shopping needs through conversation and relating product details to dialogues for better engagement. The successful development has enabled Farfetch to deliver a high-touch, personalized experience at scale, meeting modern consumers' expectations and reinforcing its market leadership.
AI-powered Psychological Coaching Platform for Healthcare
https://www.novii.ai/I spearheaded the development of Novii, an AI-powered psychological coaching platform advancing personalized healthcare through secure, continuous, and data-driven guidance.
Defined the platform architecture using FastAPI and LangGraph, aligning tech decisions with clinical data integrity and long-term patient engagement goals.
Directed multi-agent orchestration via LangGraph and Groq LLMs, leveraging Pydantic AI to enforce structured reasoning and reliable data validation across workflows.
Established a durable memory layer on Supabase (PostgreSQL with Row-Level Security), ensuring privacy, auditability, and continuity in user progress tracking.
Enhanced transparency and reliability by integrating Logfire for real-time observability and automated test pipelines for confident, high-quality releases.
Scaled deployment with Docker, GitHub Actions, and Amazon ECR, delivering reproducible builds and streamlined operational workflows.
Introduced a minimal Next.js interface to accelerate clinician adoption and gather early insights from pilot users.
Result: Delivered a production-ready, privacy-first coaching platform now in user testing: enabling traceable, psychologically grounded AI support for healthcare professionals and patients.
Defined the platform architecture using FastAPI and LangGraph, aligning tech decisions with clinical data integrity and long-term patient engagement goals.
Directed multi-agent orchestration via LangGraph and Groq LLMs, leveraging Pydantic AI to enforce structured reasoning and reliable data validation across workflows.
Established a durable memory layer on Supabase (PostgreSQL with Row-Level Security), ensuring privacy, auditability, and continuity in user progress tracking.
Enhanced transparency and reliability by integrating Logfire for real-time observability and automated test pipelines for confident, high-quality releases.
Scaled deployment with Docker, GitHub Actions, and Amazon ECR, delivering reproducible builds and streamlined operational workflows.
Introduced a minimal Next.js interface to accelerate clinician adoption and gather early insights from pilot users.
Result: Delivered a production-ready, privacy-first coaching platform now in user testing: enabling traceable, psychologically grounded AI support for healthcare professionals and patients.
Education
2008 - 2012
PhD in Mathematics and Computer Science
University of Porto - Porto, Portugal
2007 - 2008
Master's Degree in Informatics and Applied Mathematics
University of Porto - Porto, Portugal
Skills
Libraries/APIs
Scikit-learn, XGBoost, Pandas, PyTorch, PySpark, Pydantic
Tools
PyCharm, Jira, LaTeX, CircleCI, GitHub, BigQuery, Zoom, AWS Command Line Interface (CLI), Whisper, Jenkins, Git, Makefile, Docker Compose, Sentry
Languages
Python, Bash, Bash Script, SQL
Frameworks
LightGBM, LangGraph, C4 Model, Flask, Streamlit
Platforms
Linux, Docker, Databricks, Google Cloud Platform (GCP), Amazon Web Services (AWS)
Storage
Elasticsearch, MongoDB, PostgreSQL, Redis
Paradigms
Test-driven Development (TDD), REST, Management
Other
Artificial Intelligence (AI), Machine Learning, Computer Vision, DVC, Light LLMs, Large Language Model Operations (LLMOps), Large Language Models (LLMs), Data Augmentation, Recommendation Systems, Search, Information Retrieval, Learning to Rank, Deep Learning, Team Leadership, Conversational AI, Complex Problem Solving, Creative Problem Solving, Strategy, Lean Project Management, Open-source LLMs, Machine Learning Operations (MLOps), Scalable Web Services, OpenAI GPT-3 API, API Integration, Chatbots, Software Architecture, Data Science, OpenAI GPT-4 API, Fine-tuning, Embeddings from Language Models (ELMo), Semantic Search, eCommerce, Technical Leadership, Predictive Modeling, LangChain, Agentic AI, AI Chatbots, Predictive Analytics, Prompt Engineering, Solution Architecture, Image Analysis, AI Assistants, Troubleshooting, Vector Search, Clustering, Natural Language Generation (NLG), Retrieval-augmented Generation (RAG), APIs, Generative Pre-trained Transformers (GPT), Hugging Face, Generative Artificial Intelligence (GenAI), Startups, Vector Databases, AI Agents, Full-stack, Discord, GitHub Actions, Optimization, Generative Pre-trained Transformer 4 (GPT-4), OpenAI, Sphinx, PlantUML, Transformer Models, PT-Ranking, Octopus Deploy, Natural Language Processing (NLP), Communication, Remote Team Leadership, FastAPI, Personalization, MLflow, Data Engineering, Computer Vision Algorithms, Statistics, Bayesian Statistics, Naive Bayes, Bayesian Inference & Modeling, Algorithms, Stochastic Process, Numerical Methods, Numerical Analysis, Applications, Linear Optimization, Web Scraping, Taxonomy, UNet, Diffusion-based AI Models, Qdrant, Supabase
Collaboration That Works
How to Work with Toptal
Toptal matches you directly with global industry experts from our network in hours—not weeks or months.
1
Share your needs
Discuss your requirements and refine your scope in a call with a Toptal domain expert.
2
Choose your talent
Get a short list of expertly matched talent within 24 hours to review, interview, and choose from.
3
Start your risk-free talent trial
Work with your chosen talent on a trial basis for up to two weeks. Pay only if you decide to hire them.
Top talent is in high demand.
Start hiring