Software & Machine Learning Engineer · Würzburg, Germany

Junaid
Ahmad

I build machine learning systems and software that hold up in practice: LLM applications with validated, source-grounded outputs, deep learning models for orbital mechanics and weather forecasting, and production code in Python and C#.

2,900periodic cislunar orbits analysed with unsupervised learning
35%lower crash and ANR rates in released mobile apps
60%fewer critical bugs on a live game through debugging and optimisation
20%longer average sessions on the same live game
01

About

I care about building AI systems whose results can be checked, explained and trusted.

I'm a software engineer and M.Sc. Computer Science student at Julius-Maximilians-Universität Würzburg. My work covers deep learning, fine-tuning with PEFT/LoRA, and agentic AI workflows built with LangChain and LangGraph.

I'm particularly interested in multi-agent systems, retrieval-augmented generation (RAG) and knowledge-based applications.

Before my master's I worked as a software engineer at The Game Storm Studios in Lahore, building mobile, VR and multiplayer applications in Unity and C#.

Location
Würzburg, Germany · open to relocation
Languages
English (C1) · German (B1)
Focus
Machine learning, LLMs, software engineering
02

Experience

  1. 10.2025 — 09.2026

    Machine Learning Research Intern

    Julius-Maximilians-Universität Würzburg · Würzburg, Germany

    Cislunar Orbital Feature Extraction Apr 2026 — Sep 2026

    • Applied unsupervised learning to about 2,900 periodic cislunar orbits using K-Means, DBSCAN, Mean Shift and Affinity Propagation.
    • Engineered orbital features covering state, energy, lunar geometry, orbital motion and trajectory curvature.
    • Evaluated how feature selection affects the identification of distinct orbital families.

    State Vector Uncertainty Propagation Oct 2025 — Mar 2026

    • Developed an MLP surrogate model that approximates Monte Carlo orbital uncertainty propagation at a fraction of the computational cost.
    • Prepared datasets of simulated cislunar trajectories and evaluated the model against Monte Carlo results.
    • Analysed performance across trajectory families and the limits of purely data-driven uncertainty propagation.
    • Python
    • Scikit-learn
    • Pandas
    • NumPy
    • Matplotlib
  2. 04.2025 — 09.2025

    Machine Learning for Time Series

    Julius-Maximilians-Universität Würzburg · ECMWF AI Weather Quest
    • Implemented and trained the Circular Transformer (CirT) in PyTorch Lightning for subseasonal weather forecasting.
    • Ran large-scale training jobs on the Julius 2 HPC cluster with SLURM to process the ERA5 reanalysis dataset.
    • Applied parameter-efficient fine-tuning (PEFT) with LoRA to adapt a model to a task-specific application with far fewer trainable parameters.
    • PyTorch
    • PyTorch Lightning
    • SLURM
    • LoRA
    • Julia
  3. 10.2024 — 04.2025

    Student Research Assistant

    FAU Erlangen-Nürnberg · Nuremberg, Germany
    • Increased average session length by about 20% on a live Roblox game.
    • Cut the critical bug count by about 60% through systematic debugging and targeted performance optimisation.
    • Managed source control and the CI/CD build pipeline with Git.
    • Lua
    • Roblox
    • Blender
    • Maya
    • Git
  4. 07.2022 — 03.2024

    Software Engineer

    The Game Storm Studios · Lahore, Pakistan
    • Reduced ANR and crash rates by about 35% across five released titles, measured with Firebase Crashlytics.
    • Built interactive VR and 3D simulation applications in Unity (C#) with real-time multiplayer networking.
    • Structured code with MVC, Observer and Command design patterns.
    • C#
    • Unity
    • VR / Oculus
    • Multiplayer
    • Firebase
03

Projects

AI hackathon · Generative AI

SmartExam

An LLM-based document pipeline that converts DOCX exams into Markdown and back, including comments, and splits them into individual tasks. Output validation and source grounding reduce hallucinations and keep every generated result traceable.

  • LLMs
  • Python
  • Document parsing
  • API integration
View on GitHub
Transformers · HPC

Subseasonal weather forecasting

Training and evaluating the Circular Transformer on ERA5 reanalysis data for 3 to 6 week temperature forecasts, run on the Julius 2 cluster for ECMWF's AI Weather Quest.

  • CirT
  • ERA5
  • PyTorch Lightning
  • SLURM
View on GitHub
Unsupervised learning

Cislunar orbit families

Clustering about 2,900 periodic Earth–Moon orbits and measuring how the choice of engineered features changes which families emerge.

  • K-Means
  • DBSCAN
  • Mean Shift
  • Feature engineering
Deep learning · Surrogate models

Fast uncertainty propagation

An MLP surrogate that replaces expensive Monte Carlo runs when propagating state-vector uncertainty along cislunar trajectories, with an analysis of where it breaks down.

  • MLP
  • Monte Carlo
  • Model evaluation
Software engineering · VR

Mobile, VR and multiplayer apps

Mobile games, VR and 3D simulations with real-time multiplayer networking, built in Unity and C# at The Game Storm Studios.

  • Unity
  • C#
  • Oculus
  • Firebase
04

Skills

Programming

  • Python
  • C#
  • C++
  • Lua

Machine learning

  • Deep Learning
  • Transformers
  • Neural Networks (MLP)
  • Unsupervised Learning
  • Clustering
  • Feature Engineering
  • Model Evaluation

Generative AI & LLMs

  • Large Language Models
  • Agentic AI
  • Prompt Engineering
  • Fine-Tuning
  • PEFT / LoRA
  • Source Grounding
  • Hallucination Mitigation

Frameworks

  • PyTorch
  • PyTorch Lightning
  • LangChain
  • LangGraph
  • Scikit-learn
  • NumPy
  • Pandas
  • Matplotlib

Tools & methods

  • Git
  • CI/CD
  • HPC (SLURM)
  • API Integration
  • Unity
  • Firebase
05

Education

04.2024 — Present

M.Sc. Computer Science

Julius-Maximilians-Universität Würzburg, Germany

10.2018 — 07.2022

B.Sc. Computer Science

University of Central Punjab, Lahore, Pakistan

Certifications
  • Building Agent SkillsAnthropic
  • AI Fluency: Framework & FoundationsAnthropic
  • AWS for SAPAWS
Languages
  • EnglishC1
  • GermanB1
06

Get in touch

I'm happy to talk about roles, collaborations or projects in machine learning and software engineering.