Pelumi Victor Aderinto

Research interests

• Multimodal Machine Learning and Vision-Language Models
• Generative AI and Large Language Models
• Medical Imaging and AI for Healthcare
• AI for Science and Scientific Machine Learning
• Structured and Interpretable Machine Learning
• Reasoning and Representation Learning

Education

1. 2023 – 2024

   MSc in AI for Science

   African Institute for Mathematical Sciences (AIMS) South Africa · Affiliated university: University of Cape Town · Google DeepMind Scholar · Overall Distinction

   Graduated with Overall Distinction; named Salutatorian and received the Academic Excellence Award at the AIMS recognition of awards ceremony (2024).

   • Thesis: “Automatic Radiology Report Generation from 3D CT Scans Using GPT-2 Pre-Trained Large Language Model (LLM)”, in collaboration with the University of the Western Cape; published in Springer LNICST vol. 697 (PAAISS 2025).
   • Google DeepMind Scholarship for the AI for Science programme (2023 – 2024).

2. 2016 – 2022

   BSc (Hons) Electronic and Electrical Engineering

   Obafemi Awolowo University (OAU), Osun State, Nigeria · Second Class Upper Division · top 5% of class

   Final CGPA 4.46/5.0 (89%), major GPA 5.0/5.0. Thesis: “Design and Construction of a Bidirectional Net Energy Meter with Web and SMS Notification”.

   • 2nd Overall Best Student in Control Engineering, Electronic and Electrical Engineering (2021).
   • Led undergraduate design groups: a strain-gauge weight sensor, an SVM-based anomaly detector for over- and under-voltage in power systems, and an automatic irrigation system.

Research and professional experience

1. June 2025 – Present

   AI Research Engineer

   IQAI · Embodied AI & Real-Time LLM Systems

   Real-time speech-to-speech AI pipelines, LLM reasoning and robot control for an embodied humanoid AI system, with responsibility for latency and reliability.

   • Built real-time speech-to-speech pipelines integrating streaming speech recognition, LLM reasoning and text-to-speech for an embodied humanoid AI system.
   • Benchmarked GPT-4o, Gemini, Groq, ElevenLabs and Cartesia on latency and reliability, and optimised voice-activity detection to reduce conversational turn-detection latency.
   • Developed a Python robot control framework and action API supporting locomotion, gaze, end-effector control and real-time interpolated joint control across simulation and physical hardware.
   • Designed and deployed an event-driven system connecting on-chain events to autonomous robot behaviours, executing 98 action sequences over 10 days without failure.

2. November 2024 – May 2025

   AI Research Engineer Fellow

   InstaDeep · InstaGeo team (AI for Social Good) · Young African AI Research Fellowship

   Scientific machine learning for Earth observation: physics-informed retrieval of vegetation biophysical parameters from Sentinel-2 imagery using the PROSAIL radiative transfer model.

   • Developed a self-supervised framework integrating a PyTorch implementation of PROSAIL to train a ResNet-MLP for canopy parameter retrieval (leaf area index, canopy chlorophyll content, canopy water content).
   • Reduced memory overhead by 60% through chunk-based data generation (batch sizes up to 5k–10k) and optimised GPU usage.
   • Implemented parameter bounding with scaled sigmoid transforms to prevent physically invalid reflectance in the radiative transfer simulations.
   • Validated model performance against Sentinel-2 in-situ measurements of LAI and CCC; co-developed the physics-informed Transformer-VAE reported in arXiv:2511.10387 with Prince Mensah, Ibrahim Salihu Yusuf and Arnu Pretorius.

3. January 2024 – November 2024

   AI Researcher, Radiology NLP

   AIMS South Africa / University of the Western Cape collaboration

   Thesis research on automatic radiology report generation from 3D chest CT, with Mehrdad Ghaziasgar (UWC), published in Springer LNICST (PAAISS 2025) and awarded 2nd Best Paper.

   • Designed a multimodal architecture combining a RadImageNet-pretrained ResNet-50 and GPT-2 to generate expert-level radiology reports from 3D chest CT scans.
   • Devised a 3D-aware Axial-Coronal-Sagittal (ACS) convolution strategy to adapt 2D CNNs to volumetric CT data, and integrated cross-attention layers into GPT-2 under a single 24 GB GPU budget.
   • Achieved BLEU-1 of 0.403 and perplexity of 2.34 on CT-RATE while training on about 20% of the dataset (~11.6k volumes), approaching the state of the art (CT2Rep) under a 3 TB compute budget.
4. Graduate Assistant

   Obafemi Awolowo University · Department of Electronic and Electrical Engineering

   Taught, supervised and graded more than 20 undergraduates in assignments and tests for EEE301, Microelectronics Devices and Circuits I, under the supervision of Dr. O. O. Ilori.

Publications

Awards and scholarships

1. October 2025

   Award

   2nd Best Paper Award

   4th Pan-African Artificial Intelligence and Smart Systems Conference (PAAISS 2025)

   PAAISS 2025, Kwame Nkrumah University of Science and Technology, Kumasi, Ghana

2. 2024

   Fellowship

   Young African AI Research Fellowship

   InstaDeep

   InstaDeep

3. 2024

   Award

   Salutatorian and Academic Excellence Award

   MSc in AI for Science, graduated with Overall Distinction

   African Institute for Mathematical Sciences (AIMS) South Africa

4. 2023 – 2024

   Scholarship

   Google DeepMind Scholar Award

   AI for Science Master's programme

   African Institute for Mathematical Sciences (AIMS) South Africa

5. 2021

   Award

   2nd Overall Best Student in Control Engineering

   Electronic and Electrical Engineering

   Obafemi Awolowo University, Osun State, Nigeria

6. 2015

   Award

   State Best 10, Senior Mathematics Level

   Nigerian Mathematics & Sciences Olympiad

   Nigerian Mathematics & Sciences Olympiad

7. 2015

   Award

   Overall Best Student in Mathematics

   Class of 2015

   Oladipo Alayande School of Science (OASOS)

Talks and tutorials

1. 2 August 2026

   Talk

   The Math Behind Neural Networks

   Tutorial, Deep Learning Indaba 2026

   Deep Learning Indaba 2026 · Pan-Atlantic University, Lagos, Nigeria · Co-presented with Catherine Essuman

Languages

• English (fluent)
• Yoruba (native)

Technical expertise

Machine learning and NLP

• PyTorch and Hugging Face Transformers
• LLM fine-tuning (QLoRA/PEFT, Unsloth); quantisation and on-device deployment (GGUF, llama.cpp)
• Retrieval-augmented generation (FAISS, sqlite-vec, embedding models)
• Multimodal vision–language modelling
• Real-time speech–LLM pipelines (streaming STT/TTS, Pipecat)
• Self-supervised and physics-informed machine learning
• Evaluation and benchmarking (BLEU/ROUGE, mathematics benchmarks, latency and TTFB profiling)

Engineering

• Python
• Dataset creation and curation; data engineering and ETL
• Cloud and constrained-compute training (Google Cloud, Colab, Kaggle)
• Robot control frameworks and action APIs across simulation and hardware
• Git and LaTeX

Research and communication

• Academic research and technical writing
• Teaching and tutorial development
• Multilingual data curation and evaluation

Professional memberships

• Black in AI
• Graduate Member, Nigerian Society of Engineers (NSE)

Certifications

• Neural Networks and Deep Learning (DeepLearning.AI)
• Improving Deep Neural Networks: Hyperparameter Tuning, Regularization and Optimization (DeepLearning.AI)