🚀 Flagship Systems

Five projects that best represent the architectural approach: building infrastructure that compounds, not tools that expire.

Battery Simulation Framework + PINN Degradation

Volvo Trucks R&D · 2022
Problem: Battery simulation was time-consuming, as it was run on full-vehicle simulation framework
System built: Standalone simulation environment with BMS as a first-class citizen.
Core Impact: Integrated a PINN degradation model running at ~90% accuracy in seconds on mobile

Current Limits Generator

A123 Systems · 2025
Problem: current limits were being computed manually, inconsistently, and conservatively across OEM programmes.
System built: physics-based Python tool generating aging-aware current envelopes covering plating onset, side reactions, thermal limits, and thermal runaway margin - parameterised, OEM-deliverable, in production use.
Core Impact: Standardized and physics-oriented methodology to calculate dynamic current limits, which are used as input for battery simulations, particularly powermaps calculation.

Battery Thermal Model Configurator \

Mercedes-Benz R&D · 2019-2021 · tool
Problem: To build an accurate and faster battery thermal model coupled with cooling plates/thermal management system.
System built: Full fledged matlab based tool, which couples CFD cooling channel results to a reduced-order thermal ROM - generates a fully-parameterised Simulink thermal model.
Core Impact: 60% reduction in model development time. | Users - R&D engineers in Mercedes, Stuttgart

Codebase Indexer

Self-initiated · 2025–present
Problem: AI coding agents generate code without understanding the codebase they’re editing.
System built: structural repository understanding layer - dependency graphs, importance scoring, semantic retrieval - that gives a local LLM a map before it touches a line.
Core Impact: codebase mapping problem resolved. No hallucinations. Json output file can be used by AI agents.
GitHub →

Hybrid Code Analyzer

Self-initiated · 2025
Problem: static analysis misses runtime failures; dynamic analysis misses structural importance.
System built: combined pipeline that correlates runtime failures with structural centrality - finding what neither approach alone can detect.
GitHub →

OpenFOAM Battery Simulator

Self-initiated · Open source · 2025
Problem: CFD-based battery simulation is powerful and routinely impractical. System built: Python interface layer that converts OpenFOAM from an expert-only C++ tool into a callable function - enabling parameter sweeps, optimisation loops, and AI integration. GitHub →


🔋 Battery & Electrochemistry

The core domain - electrochemical modelling, thermal simulation, ageing, and system-level battery behaviour. Work spans cell characterisation through pack-level simulation across five companies.

  • Battery Modelling
    Cross-company · 2015–present · analysis
    ECM vs DFN vs data-driven - what each gets wrong, and why hybrids are the real answer.

  • Thermal Management System Modeling
    Mercedes-Benz R&D · 2015�?021 · tool
    Building the full EV TMS from scratch - battery, inverter, DC-DC, coolant loop, HVAC - integrated into the in-house vehicle simulation framework.

  • DC Box Thermal Modelling
    Mercedes-Benz R&D · 2016 · analysis
    The component nobody thinks about until it fails. 96% experimental accuracy; design margin identified before hardware.

  • Charging Time Modelling
    Mercedes-Benz R&D · 2017�?018 · analysis
    CC-CV physics, C-rate tradeoffs, and why fast charging is a thermal problem as much as an electrochemical one.

  • Battery Simulation Framework
    Volvo Trucks R&D · 2022 · tool
    Standalone simulation environment with BMS-as-first-class-citizen, plus a PINN degradation model built for mobile inference.

  • PINN Battery Degradation
    Volvo Trucks R&D · 2022 · method
    Physics-Informed Neural Network for battery degradation - physics-constrained training that prevents physically impossible predictions. Built for real-time aging inference on mobile hardware.

  • Virtual Cell Scaling
    A123 Systems · 2025–present · tool
    Physics-based scaling across capacity, form factor, and chemistry - for when you don’t have the cell you need.

  • OpenFOAM Battery Simulator
    Self-initiated · 2025 · open source · tool
    Python interface layer turning OpenFOAM CFD into a callable function - enabling parameter sweeps, optimisation, and AI integration. Full writeup →


🛠�?Engineering Tools Built

Tools that exist, are used, and reduce the cost of doing the next project. Mix of OEM-commissioned and self-initiated.

  • Battery Thermal Model Configurator
    Mercedes-Benz R&D · 2019�?020 · tool
    60% reduction in thermal model development time. Generates parameterised CFD-thermal ROM for HV battery systems.

  • Model Validation Framework
    Mercedes-Benz R&D · 2018�?020 · tool
    Automated pipeline: fetch test data + battery model �?run simulations �?validate �?upload report.

  • Current Limits Generator
    A123 Systems · 2025 · tool
    Physics-based current envelope covering lithium plating onset, side reactions, thermal limits, and thermal runaway margin - aging-aware, OEM-deliverable.

  • Pack Cost Estimator
    A123 Systems · 2024 · tool
    RFQ/pre-RFQ phase commercial tool - cell selection through pack concept cost in a single run.

  • Front Fender Drag Optimization
    Hero MotoCorp · 2014 · method + tool
    ANN surrogate + Firefly algorithm. 12% drag reduction, 10× speedup versus CFD-in-loop. The method generalises.


🤖 AI & Code Intelligence

Self-initiated - built outside the day job to solve problems the existing toolchain couldn’t. Runs on local LLMs (LM Studio) for GDPR-compliant on-premise deployment - relevant for the German automotive context where sending engineering data to external APIs is not an option.

  • Battery Expert AI
    Self-initiated · 2025–present · tool
    Local, on-premise AI assistant for battery engineering - LoRA fine-tuned on domain knowledge, RAG over electrochemistry papers and test data, with mathematical equation reasoning for PDEs and electrochemical models.

  • AI Systems Overview
    Self-initiated · 2025–present · overview
    The design philosophy connecting these tools: structure before generation, understanding before output.

  • Codebase Indexer
    Self-initiated · 2025–present · tool
    FAISS-based pre-filter navigation layer for local LLMs. Dependency graph, importance scoring, structural query layer. The difference between an agent that codes and one that understands.

  • Hybrid Code Analyzer
    Self-initiated · 2025 · tool
    Static + dynamic analysis pipeline. Finds what neither alone can - runtime failures invisible to AST analysis, structural issues invisible to execution tracing.

  • AI-Assisted Simulation Debugger
    Self-initiated · 2026 · concept
    Exploratory: treating simulation failures as propagating structural inconsistencies rather than isolated errors - AI-assisted root cause reasoning for engineering simulations.

  • Physics-Informed Neural Network - Battery Degradation
    Volvo Trucks R&D · 2022 · method (cross-listed: Battery section)
    Where domain physics meets ML - PINN-constrained degradation model prevents physically impossible predictions while enabling real-time inference.

  • Alchemist
    Self-initiated · 2025 · tool
    Indexes past repositories, extracts architectural patterns, suggests new combinations. Synthesis over generation.

  • Simple RAG Chatbot
    Self-initiated · 2025 · tool
    First local RAG experiment - document retrieval, context injection, local LLM inference. The seed of the current toolchain.


⚙️ Mechanics, CFD & Dynamics

The foundation - structural analysis, contact mechanics, vibration, and fluid dynamics from the first decade of the career. The physics intuition from this work keeps appearing in battery and AI contexts.

  • Tire Modelling
    IIT Kanpur M.Tech · 2010�?011 · analysis
    Standing waves, critical velocity, contact mechanics. The thesis that established the pattern: rotate a structure and everything changes.

  • Piston Rings
    TVS Motor Company · 2012�?013 · analysis
    Three simultaneous contact regimes, elastohydrodynamic lubrication, ring flutter. Where the interesting failure always lives at the boundary.

  • Magnesium Alloy Wheel Design
    TVS Motor Company · 2011�?012 · method + analysis
    Topology optimisation, fatigue under realistic load spectra, thermal analysis. Material substitution is a system problem, not a materials problem.

  • Engine Mounts: Designing Something to Fail Softly
    Hero MotoCorp · 2013�?015 · method
    NVH optimisation through stiffness design. Hybrid GA + Nelder-Mead, 15�?0% cabin vibration reduction, verified on dynamometer.

  • Radar Modeling for Autonomous Driving
    Mercedes-Benz R&D · 2016�?017 · analysis
    What sensors actually see - beam patterns, clutter, Kalman filter state estimation, and why edge cases aren’t edge cases.

  • CFD Fire Modelling
    Victoria University, Melbourne · 2009 · method
    Material property estimation by running the physics model backwards. GA + Nelder-Mead, cross-validated, published.


📄 Career �?timeline, education, patents →

🔬 Beyond the work �?Research Interests → · Dynamics, contact mechanics, PINNs, evolutionary algorithms, and the patterns that keep appearing across domains that look unrelated on the surface.