Portrait of Philip Obhenimen John

Combustion CFD · Reduced-Order Modeling · Scientific ML & AI

Philip Obhenimen John

I build physics-based and data-driven models of turbulent reacting flows: CFD of liquid-fueled gas-turbine combustors, reactor-network reduced-order models that run more than 1,000× faster, and the machine learning, deep learning, and AI that tie them together.

Explore my research Download CV

Open to full-time roles from Spring 2027 in scientific machine learning, combustion and CFD modeling, and AI for simulation. See my CV or email me.

01 — PhD Research · DMO Lab, LSU

From high-fidelity CFD to real-time combustion models

CFD is the most trusted way to predict combustor emissions, but each operating point costs thousands of CPU-hours. My dissertation builds a pipeline that keeps the physics and drops the cost.

  1. 01

    High-fidelity simulation

    Combustion CFD

    RANS and LES of swirl-stabilized, liquid-fueled gas-turbine combustors with Lagrangian sprays, tabulated chemistry, and adaptive mesh refinement on HPC.

    ~3.2M cells · 10 h on 256 cores

  2. 02

    Reduced-order modeling

    Reactor networks

    Clustering turns millions of CFD cells into a handful of chemical reactors with detailed Jet-A kinetics, including custom spray-evaporation reactors I developed.

    O(10) s on 1 core · >1,000× faster

  3. 03

    Machine & deep learning

    ML-augmented models

    RL agents build the networks automatically, multi-fidelity Gaussian processes fuse cheap network data with scarce CFD, and neural networks predict NOx.

    CFD accuracy with half the CFD runs

  4. 04

    Generative AI

    Diffusion models

    A diffusion model that reconstructs high-resolution LES flow and species fields from reactor-network output, one of the final projects of my PhD.

    Goal: LES detail at network cost

My large-eddy simulation (top, 18,400 CPU-hours) next to my reactor-network model of the same combustor (bottom, 0.09 CPU-hours).
2,300×
LFRN speed-up over 256-core CFD
2×
CFD data efficiency with multi-fidelity GPs
0.75→0.92
Mean R² at 7 CFD points, GP vs. multi-fidelity
2
First-author journal papers

Full research details, methods & videos →

02 — Expertise

What I do, and the proof

Every skill here is backed by peer-reviewed research or shipped software.

Primary focus

Combustion CFD & Reduced-Order Modeling

I simulate multiphase turbulent reacting flows and compress them into reduced-order models fast enough for emissions prediction and design-space exploration.

  • Liquid-Fueled Reactor Network: to our knowledge, the first coupled reactor-network framework for liquid-fueled combustors, with reactors for spray breakup, droplet heating, and evaporation. It was validated against CFD across inlet temperatures of 300–500 K and a range of fuel flow rates (Combustion and Flame).
  • Combustor CFD in CONVERGE: RANS k-ε, flamelet-generated manifold chemistry with a presumed β-PDF, Lagrangian sprays (TAB breakup, Frossling evaporation), and AMR.
  • Detailed kinetics: Cantera with the HyChem Jet-A mechanism (119 species, 841 reactions), resolving thermal, prompt, and N2O-pathway NOx.
  • CONVERGE
  • ANSYS Fluent
  • OpenFOAM
  • Cantera
  • OpenSMOKE++
  • Sprays
  • Turbulent combustion
  • NOx
  • Lean blowout

Data-efficient modeling

Machine Learning

  • Multi-fidelity Gaussian processes (NARGP) that fuse 400 reactor-network runs with 35 CFD simulations, using half the CFD for the same accuracy.
  • Reinforcement learning, k-means clustering, and dimensionality reduction to compress CFD fields.
  • Gaussian processes
  • Multi-fidelity
  • RL
  • emukit
  • scikit-learn

Neural networks

Deep Learning

  • Neural networks that predict NOx emissions from reduced-order model outputs.
  • Deep generative models trained on HPC simulation data through automated pipelines.
  • PyTorch
  • TensorFlow
  • Surrogates

Generative AI

Diffusion Models

  • Ongoing PhD project: a diffusion model that reconstructs high-resolution LES flow and species fields from reactor-network data.
  • Diffusion
  • Generative modeling
  • Scientific ML

Agentic AI

AI Agent Engineering

  • RL agents that make sequential decisions to build reactor-network models automatically, improving lean-blowout prediction (Proc. Combustion Institute).
  • MyStudySidekick: a production agentic backend where LLM agents use verification tools and skills to check study material before students see it, on top of a multimodal RAG pipeline.
  • Poscholars: AI chat over students' own uploaded course files, with page selection and image extraction for grounded answers, plus AI-generated flashcards, serving 40,000+ students.
  • Mitrawriter: agentic workflows for scientists, including cross-paper chat, citation-aware LaTeX writing, code-to-paper integration, and thesis assembly.
  • Agentic workflows
  • RAG
  • LLM APIs
  • Embeddings

Scientific computing

HPC & Scientific Software

  • Automated, parallel Python pipelines on LSU's SuperMike-3 cluster linking simulation, data extraction, model construction, and ML training.
  • Large parametric NOx and lean-blowout campaigns; flamelet tables generated with OpenSMOKE++.
  • Native-code optimization of performance-critical paths in C++, Swift, and Kotlin.
  • MPI
  • OpenMP
  • SLURM
  • C++
  • Python
  • Linux
  • ParaView

Technical toolkit

CFD & combustion
CONVERGE, ANSYS Fluent, OpenFOAM, Cantera, OpenSMOKE++, ParaView
Physics
Multiphase sprays, turbulent combustion, chemical kinetics, FGM, NOx, lean blowout
ML & AI
PyTorch, TensorFlow, scikit-learn, emukit; diffusion models, RL, Gaussian processes, neural networks
Agentic AI
Agentic workflows, tool use, verification tools, skills, RAG, embeddings, LLM APIs (Gemini, OpenAI), Whisper
HPC & tooling
MPI, OpenMP, SLURM, Linux, parallel Python pipelines, Git, CI/CD
Languages
Python, C++, Go, TypeScript, Rust, Swift, Kotlin, MATLAB, SQL

03 — Publications

Papers

  1. A Reinforcement-Learning-Augmented Liquid-Fueled Reactor Network Model for Predicting Lean Blowout in Gas Turbine Combustors
    John, P., Ikponmwoba, E., Pal, P., & Owoyele, O. (2026). Proceedings of the Combustion Institute, 42, 106463. doi:10.1016/j.proci.2026.106463
  2. A Liquid-Fueled Reactor Network Model for Enhanced NOx Prediction in Gas Turbine Combustors
    John, P., Chandrasekhar, H., Saha, S., & Owoyele, O. (2026). Combustion and Flame, 292, 115199. doi:10.1016/j.combustflame.2026.115199
  3. Multi-Fidelity Gaussian-Process Surrogates for Data-Efficient Prediction of Gas-Turbine Combustor Emissions
    John, P., & Owoyele, O. In preparation.

Peer reviewer for Combustion and Flame and the Journal of Energy Resources Technology.

04 — Software & Products

Things I've built and shipped

Beyond the lab, I design and ship production software end to end, from agentic AI tools to mobile apps used by more than 40,000 students.

  1. 001

    MyStudySidekick

    An agentic AI study platform for STEM students, live on iOS, Android, web, and desktop.

    AI agents · Go · React Native
  2. 002

    Poscholars

    A mobile learning app with AI chat serving 40,000+ students in Nigeria on the App Store and Google Play.

    NestJS · Swift · Kotlin
  3. 003

    Animal Game

    An animal quiz game with 495 animals and six game modes, published by Poscholars.

    iOS · Android · Game
  4. 004

    Mitrawriter

    An agentic desktop research environment that reads, writes, and assembles papers, slides, and theses alongside scientists.

    AI agents · RAG · Beta Dec 2026

05 — About

Simulate.
Learn.
Ship.

I'm a PhD candidate in Mechanical Engineering at Louisiana State University, working in the Data-driven Modeling and Optimization (DMO) Lab and advised by Dr. Ope Owoyele. My dissertation, Liquid-Fueled Reactor Network Modeling for Gas Turbine Combustors, combines combustion CFD, reduced-order modeling, and machine learning to make high-fidelity simulation fast enough for design loops.

I'm equally at home writing production software. I've built and shipped AI and mobile apps used by tens of thousands of people, which shapes how I build research code: automated, reproducible, and ready to scale on HPC.

Researcher

Combustion CFD · Reactor networks · Sprays · Chemical kinetics · HPC

ML & AI

Diffusion models · Deep learning · Gaussian processes · RL · Agentic AI

Builder

MyStudySidekick · Poscholars · Mitrawriter · 40k+ students reached

Education

  1. 2022 – Spring 2027 (expected)
    Ph.D., Mechanical Engineering, Louisiana State University · GPA 3.95
    Coursework: Machine Learning for Engineers, High-Performance Scientific Computing, Numerical Methods
  2. 2014 – 2019
    B.Eng., Mechanical Engineering, University of Benin
    CGPA 4.83/5.0, First Class Honours, Best Graduating Student

Experience

  1. 2022 – Present
    Graduate Research Assistant, DMO Lab, LSU
  2. 2022 – Present
    Teaching Assistant, Thermodynamics, Fluid Mechanics, Heat Transfer
    Mentoring undergraduates on research and senior design
  3. 2025 – Present
    Software Engineer, MyStudySidekick
  4. 2021 – Present
    Co-founder & Lead Developer, Poscholars
  5. 2018
    AI Research Intern, Xigma Input and Output Technology
    ML models for engineering prediction; numerical PDE solvers in Python and MATLAB

06 — News

Recent

  1. Mitrawriter, an agentic research environment for scientists, opens its public beta.

  2. Liquid-fueled reactor network paper published in Combustion and Flame, vol. 292.

  3. RL-augmented reactor network paper published in Proceedings of the Combustion Institute, vol. 42.

  4. Launched MyStudySidekick on iOS, Android, web, and desktop.

  5. The Poscholars app launches on the App Store.

  6. Started my PhD at Louisiana State University.

  7. Co-founded Poscholars, an education platform for Nigerian university students.

  8. Graduated as best graduating student, University of Benin; won the Edo State/Siemens Energy Hackathon.

  9. 2nd place team, Hult Prize Nigeria National Finals; UN Academic Impact Millennium Fellow.

07 — Contact

Let's build something together.

I'm available for full-time roles starting Spring 2027. The fastest way to reach me is email.

philipobhenimen@gmail.com LinkedIn Google Scholar

Most of my production code is proprietary. I'm happy to walk through architecture and design decisions in conversation.