CV

Olaitan Philip Olaleye

Phone: 413-835-5251 | Location: Short Hills, NJ (NY Metro Area) | Email: natialol@gmail.com | LinkedIn: https://www.linkedin.com/in/olaitan-olaleye-71233027/

🧠 Expertise

Deep Learning (Generative AI, LLM, NLP, Computer Vision, LVM, Multi-Modal, Foundation & Document AI), ML Technical Leadership, ML System Design, Model Optimization, RLHF / DPO / CRPO, Innovation & Cross-Organization Impact, Patent Generation & IP Development.

💼 Professional Background

I am a Staff-level Applied Research Scientist at Oracle Cloud where I’ve helped lead the design and build-out of globally deployed services spanning pre-trained Generative AI, Responsible AI, Object Detection, Invoice/Receipt Key-Value Extraction, Custom KV/DOCUMENT AI, Medical Imaging, as well as their Auto-ML variants. I currently lead some efforts in the Generative AI division.

Previously, I served as Tech Lead at Signify (formerly Philips) for new deep-learning applications, working on innovative services such as Lightfinder and Global Streetview Audit. I was also a Research Scientist at Amazon, focusing on ML-driven automation for employee decision making and performance management — efforts that contributed to company-wide adoption of new recommendations.

My work has resulted in over 60 patent applications (20+ granted) and multiple peer-reviewed publications. I hold a Ph.D. in Optimization, a B.Sc. in Engineering, and an MBA.

📂 Experience

Oracle Cloud Infrastructure

Staff Applied Scientist — 12/2020 – Present

• Current role (2025): Reasoning & Agentic AI Lead — spearheading launch of an end-to-end multi-task automated prompt-optimization framework, now extending to agentic use-cases. Leading state-of-the-art foundation model research and development, model reasoning, and test-time augmentation for accuracy enhancement.
• 2023–2024: Generative AI & Responsible AI Lead — delivered inference-protection services, domain-specific LLM fine-tuning (summarization, VQA, chat & generation), tool routing, prompt optimization, and multi-modal embedding model development via multi-task training (using Slurm, Axolotl, Accelerate, DeepSpeed, etc.). All services are customer-facing and revenue-generating.
• 2022–2023: Document AI Science & Tech Lead — built Invoice/Receipt/Custom Key-Value extraction, Health Insurance Card OCR models, and prototype Generative AI services for upcoming launch.
• 2020–2022: Vision & Medical Imaging Service Lead — led development of Object Detection, Image Classification, Video AI (research stage), and Medical Imaging pipelines (research stage). Oversaw the full lifecycle — from conception and data curation (terabyte-scale datasets) to architecture design, performance benchmarking, deployment, user testing, and service launch. Achieved > 20% improvement over comparable cloud offerings (GCP, AWS, Azure) in latency and/or accuracy.
• Additional responsibilities: Infrastructure & architecture design (ARM vs GPU, cluster training), auto-ML & hyperparameter search, efficient multi-cluster training, incremental & class-sampling learning, code review, hiring, mentoring, stakeholder coordination, IP/patents, system design and mid-term strategy planning.

Technologies & architectures used include variants of: Llama, DeepSeek, CodeLlama, Qwen, YOLOv5, Mask R-CNN, Detectron2, LayoutLMv2/v3, SDMGR, BERT, EfficientNet, Stable Diffusion / DALL-E, deBERTa, LlamaGuard/PromptGuard, plus training/serving frameworks like DeepSpeed, Axolotl, Llamafac, and more. Techniques included: reasoning & chain-of-thought, SFT, pre-training, DPO, RLHF, model distillation, tool-calling/agent frameworks, automatic prompt optimization, multi-modal embedding, test-time augmentation, embedding techniques, and efficient inference strategies.

Amazon — Research Scientist Lead (ML/AI)

09/2019 – 11/2020

• Led projects in statistical ML and deep learning — including generalized embedding modeling (a precursor to LLMs) and large-scale proprietary data analytics.
• Projects included internal asset modeling, company-wide asset output measurement (productivity), cost & compensation optimization, change analysis, multi-rater attribution, forecasting, and long-term productivity analysis under uncertainty.
• Developed and deployed novel methods for event attribution, promotion/attrition analysis, long-run asset optimization. Led a “Science-as-a-Service” framework and an ML feature repository that reduced generic model development times by up to 70%. Provided mentoring to junior members and had project proposals accepted for large confidential initiatives.

Signify / Philips Research North America — Research Team Lead, New Deep Learning Applications

(01/2018 or 2015) – 08/2019

• Managed a team of 5+ PhD researchers and interns working on deep-learning based Computer Vision, Generative Networks, Time-Series Analysis, and Smart Cities / Streaming Data Analytics.
• Led development of:
  • Lightfinder: computer-vision based e-commerce shopping app — achieved a jump from ~30% to ~97% top-1 accuracy in wild (real-world) tests by optimizing deep-learning models (Inception-V2/V3, ResNet, VGG, Mask RCNN, YOLO, etc.), using advanced augmentations, parameter and grid search, and resolution tuning. App deployed on iOS and Android, with strong revenue potential.
  • StreetView Audit & Sensor Data Analytics: scalable deep-learning based remote audit for outdoor sensor networks — including satellite imagery registration, scalable segmentation/detection pipelines (Mask R-CNN, Faster R-CNN, etc.), enabling large contracts.
  • Generative Adversarial Networks: proposed and built a novel multi-domain emotion-conversion GAN for unpaired data (first unpaired GAN paper accepted at Interspeech 2019). Explored object in-placement and photorealistic style transfer, and compressive sensing for IoT sensor data. Collaboration with academia (e.g. MIT).
  • Satellite Imagery Asset Localization: developed U-Net & bootstrapped sequential label-generation pipelines for aerial object detection/localization in high-resolution satellite imagery. Collaborated with academic partners (e.g. MIT) and industry (DigitalGlobe).
  • Sensor & Time-Series Analytics & Edge Computing: developed compressive-sensing and model-compression techniques (teacher-student networks) for edge deployment, achieving over 95% compression, enabling ML on constrained compute devices. Also worked on infrastructure-as-a-service pricing models for long-term sensor-based contracts.

Philips Research North America — Research Scientist, AI & Data Analytics

2015 – 2017

• Worked on predictive ML, edge analytics, time-series and cross-sensor analytics for Smart Cities / IoT sensor networks.
• Developed compressive sensing algorithms for sensor data compression and teacher-student deep-learning models for edge-deployable inference under constrained compute, enabling efficient analytics on edge devices. Achieved >95% compression.
• Developed models for infrastructure-as-a-service pricing for long-term sensor-based contracts, accounting for uncertainties in deliverables and contract structures.

University of Massachusetts, Amherst — Graduate Research Assistant (Decision Making under Uncertainty Lab)

2010 – 2015

• Research in robust long-term sequential decision-making under uncertainty: portfolio optimization, swarm-based genetic algorithms under environmental uncertainty, and novel global sensitivity analysis methods.
• Thesis work covered:
  1. Scenario analysis across non-stochastically dominated future potential trajectories.
  2. Swarm-based genetic algorithms for robust policy optimization under uncertain futures.
  3. Global sensitivity analysis techniques to tune solutions efficiently under large-scale uncertainty.

📄 Patents & Intellectual Property

• Over 60 patent applications submitted to the USPTO (20+ granted) — recognized for highest patent generation rate in the organization.
• Examples include: Dynamic Time Auto-ML model, Generative Voice Conversion, Distributed Resource Computing on constrained devices, Generative methods for product immersion, Adaptive Compressive Sensing & Cross-Modal Sampling.

📚 Selected Publications

• Olaleye, O. et al. “Monte Carlo Tree Search based Test Time Augmentation for Reasoning Enhancement”
• Afshin O; Olaleye, O. et al. “Zero-Shot Personal and Sensitive Information Detection with LLM and Search” (2025, EMNLP)
• Olaleye, O. et al. “Enterprise-Grade Prompt Injection and Responsible AI Management for LLMs” (2025, ACL ARR)
• Olaleye, O. et al. “Non-Parallel Emotional Speech Conversion” (2019, Interspeech) — first unpaired GAN based emotional speech conversion model

🎓 Education

• Ph.D. in Industrial Engineering & Operations Research, University of Massachusetts, Amherst — 2010–2015 (GPA: 3.9/4.0). NSF Scholar, Basant Nanda Scholar.
• B.Sc. in Electrical & Electronics Engineering, University of Lagos — 2002–2008. Shell Scholar, University Scholar.
• MBA (Analytics, Strategic Management & Finance), The University of Chicago, Booth School of Business — 2018–2020. Dean’s List (multiple semesters).

Other: Reviewer for ICLR and ACL; U.S. Citizen.

🤝 Volunteer & Community Engagement

• Livingston Kids Robotics Coach
• Millburn Soccer Kids Coach
• Greeting Team Member, Renaissance Church
• Board Member, MSC