Diego Marinho de Oliveira

• Built and managed up to 18 data science and ML engineering professionals across three teams, partnering with Product Managers, Heads of Product, Delivery Managers/Leads, UX/Design, Security, Legal, Consumer and GTM leaders.
• Designed and implemented the end-to-end GenAI property-search solution, including evaluation and structured outputs; led its cloud-scale deployment in one month, helping change a search approach established over 25+ years.
• Built the fine-tuning and evaluation framework for subsequent model iteration.
• Designed and implemented the AI agent framework and production agentic runtime for realestate.com.au’s first production, real-time conversational AI—including tool access, guardrails and request/response contracts—and led the design, evaluation and vendor comparison for memory.
• Designed real-time multimodal inference across golden-set creation, evaluation and serving, combining listing text and images to expose 45+ property attributes.
• Owned GenAI economics and latency through prompt caching, token-level cost tracking and performance controls, keeping production responses within 1–2 second service-level targets.
• Led the definition and rollout of AI governance controls across rate limiting, evaluation, monitoring and guardrails; partnered with Security to ensure the penetration test was completed and the solution passed its security checks.

Agentic AI and conversational experiences

• Implemented explicit tool-access, guardrail and request/response contracts within the production agentic runtime.
• Led team delivery of offline evaluation, observability, cost controls and scalable cloud integration around the consumer-facing runtime.
• Led the memory design process through evaluation and comparison across vendors.

Multimodal property enrichment

• Designed real-time multimodal inference combining listing text and images to expose 45+ property attributes for downstream search and listing experiences.
• Structured delivery across golden-set creation, model development, evaluation and serving, helping the team move the system into production with clear quality checkpoints.

Recommendations and consumer prediction

• Modernized a legacy homepage recommendation path by designing matrix-factorization training with sliding-window datasets, repeatable offline evaluation and a clear service API contract, giving the team a safer path for iterative online testing.
• Led the replacement of ageing buyer and renter segmentation models, contributing the training pipeline, feature exploration and Vertex AI approach while guiding a small team through evaluation and production transition.

• Built and led a global data science team of approximately 10, hiring and mentoring across several AI groups while developing the engineering and experimentation practices needed to move models from research into dependable customer experiences.
• Designed and implemented the end-to-end JobsDB AI search replacement—from source-data ingestion and enrichment through synonym expansion, AI-enriched lists, retrieval and behaviour-aware reranking—delivering a 14% lift in overall user interaction in a controlled Hong Kong experiment.
• Unified recommender delivery across ANZ and seven Asian markets—MY, SG, PH, TH, ID, HK and VN—using reusable retrieval, ranking, evaluation and serving foundations that produced 5–10x gains in a primary recommendation metric and a 12x peak uplift in one market.
• Re-architected high-volume email recommendations from a legacy process using event-driven pipelines and collaborative and content-based methods for daily and weekly customer journeys.
• Developed hybrid recommenders combining behavioural and content-based signals, supporting more relevant personalization across markets.
• Used 150+ online experiments alongside offline evaluation and load testing to de-risk product decisions and meet approximately 500 ms recommendation targets.

• Led a small team of data scientists and engineers delivering production recommender systems for candidates and hirers, improving candidate and hirer experiences through real-time systems at scale and delivering more than 115% uplift in a primary user engagement metric.
• Combined collaborative filtering, content-based ranking and NLP to connect data science experimentation with production experiences for both sides of the hiring marketplace.