0%
✓ Copied to clipboard!
Education Experience Projects Papers Certifications Contact

Avi
Goyal

def role() → "ML Engineer & Data Scientist"

M.Sc. Computer Science @ University of Würzburg. I build neural architectures that think, medical imaging systems that see, and AutoML pipelines that design themselves.

View Experience → Get in Touch
0%+
Val Accuracy
0+
Models Generated
0
ArXiv Papers
0×
Dataset Expanded
● LIVE INFERENCE
// hover nodes · click to re-initialize
Academic Journey

Education Roadmap

July 2018 – May 2022
B.Tech — Computer Science & Engineering
SRM Institute of Science and Technology, Chennai
Graduated with a stellar CGPA of 9.0/10 from one of India\'s top private universities. Developed foundations in algorithms, data structures, and software engineering. Completed major projects in Water Quality Classification (A+) and Flood Prediction (A+) using ML.
Machine LearningPythonFlaskData Science
🎓 CGPA: 9.0 / 10
🏛️
SRM University · 2018
SRM Institute of Science and Technology
SRM IST · Chennai, India
Julius-Maximilians-Universität Würzburg
JMU Würzburg · Germany · Est. 1402
🎯
JMU Würzburg · 2023 – Now
Oct 2023 – Present · ● Currently Enrolled
M.Sc. — Computer Science
Julius Maximilian University of Würzburg
Pursuing a Master\'s degree at one of Germany\'s oldest universities (founded 1402), specialising in Machine Learning, AutoML, and Computer Vision. Actively contributing to two ArXiv research publications while working as a Research & Data Science Assistant.
AutoMLDeep LearningLLMsResearch
📍 CGPA: 2.4 (German Scale) · In Progress
Career

Work Experience

Mar 2025 – Oct 2025
Data Science Research Assistant
Julius Maximilian University of Würzburg — AutoML & Computer Vision Lab
  • Built production AutoML using DeepSeek Coder 7B for automated neural architecture search (NAS)
  • Generated & validated ~1,900 unique neural net architectures; expanded LEMUR dataset 13×
  • Achieved 2–12% gains over hand-designed baselines across 7 benchmarks (MNIST, CIFAR-10/100, CelebA, ImageNette)
  • Pioneered Few-Shot Architecture Prompting (+11.6% on CIFAR-100); n=3 identified as optimal LLM config
  • Built sub-millisecond MD5 deduplication preventing ~100 duplicates, saving 200–300 GPU hours
PyTorchDeepSeekAutoMLLLMsNASNumPyPandas
Aug 2024 – Sept 2025
Data Analyst — Student Research Assistant
Universitätsklinikum Würzburg — AI in Medicine (Prof. Dr. Rüdiger Pryss)
  • Processed and analyzed medical OCT imaging data for detection & segmentation tasks
  • Developed ML models achieving 90%+ validation accuracy with GPU-accelerated pipelines
  • Implemented optimized training: GPU acceleration, data augmentation, batch processing
  • Supported "AI in Medicine" coursework; collaborated with physicians on data-driven research
PythonPyTorchPySparkOpenCVAlbumentationsScikit-learn
June 2022 – Mar 2023
Software Developer
PTN Event
  • Built 7+ live conference websites for Oil & Gas industry clients
  • Collaborated with BAs and product owners on functional and technical requirements
  • Participated in agile Scrum cycles ensuring on-time delivery
DjangoReactNode.jsWordPressDockerSQL
Sept 2020 – Dec 2021
Data Science & Business Analysis Intern
The Sparks Foundation
  • Built supervised ML models (regression & classification) for real-world business analytics
  • Quantitative analysis supporting data-driven decision processes
PythonScikit-learnPandasMatplotlib
Work

Featured Projects

01
🧠 AutoML · LLM · NAS
LLM-Based Neural Architecture Generation
✦ +11.6% CIFAR-100 · 1,900 architectures · 13× dataset

Production-ready AutoML using DeepSeek Coder 7B to automatically generate, validate, and benchmark neural architectures. Introduced Few-Shot Architecture Prompting and sub-ms MD5 deduplication saving hundreds of GPU hours.

PyTorchDeepSeekAutoMLNAS
arxiv.org/abs/2512.24120 →
02
🏥 Medical AI · Computer Vision
OCT Medical Imaging Segmentation Pipeline
✦ 90%+ val accuracy · GPU-accelerated · Clinical grade

Deep learning pipelines for OCT image segmentation at Universitätsklinikum Würzburg. GPU-accelerated training with advanced augmentation strategies achieving clinical-grade accuracy.

PyTorchOpenCVPySparkAlbumentations
Clinical Research · Confidential
03
💧 Classification · Flask Web App
Water Quality Classification System
✦ 7,000+ samples · Grade A+ · Multi-model comparison

End-to-end ML: EDA on 7,000+ chemical samples, feature engineering, multi-model comparison (Logistic Regression, Random Forest, SVM), Flask deployment with interactive frontend.

PythonScikit-learnFlaskSeaborn
github.com/GoyalAvi →
04
🌊 Prediction · Environmental ML
Kerala Flood Prediction — ML Pipeline
✦ 3,000+ records · Grade A+ · Ensemble benchmark

Supervised learning ensemble (Logistic Regression, Decision Tree, Random Forest, SVM) on Kerala flood dataset. Full preprocessing pipeline with comparative performance analysis.

PythonScikit-learnPandasJupyter
github.com/GoyalAvi →
05
🌐 Full Stack · Django · React
Petroleum Trade Network — Conference Platforms
✦ 7 live websites · Oil & Gas sector · Production

Designed and developed 7+ live conference platforms for the Oil & Gas sector including OGAD, Digital Twin Conference, Artificial Lift Summit, Container Shipping and Supply Chain conferences.

DjangoReactWordPressRedux
ptnevents.com →
Research

ArXiv Publications

01
Enhancing LLM-Based Neural Network Generation: Few-Shot Prompting and Efficient Validation for Automated Architecture Design

Introduces Few-Shot Architecture Prompting achieving +11.6% on CIFAR-100, systematic optimization of LLM-based NAS, and sub-millisecond MD5 deduplication. Supervised by Prof. Dr. Radu Timofte.

↗ arxiv.org/abs/2512.24120
02
NNGPT: Rethinking AutoML with Large Language Models

Novel approach to AutoML rethinking traditional neural architecture search using generative capabilities of large language models for architecture search and optimization at scale.

↗ arxiv.org/abs/2511.20333
Expertise

Technical Skills

🧠 ML & AI
Deep LearningComputer VisionNLPAutoMLNASLLMsGenerative AIPredictive Modeling
🐍 Languages
PythonGoC / C++JavaScriptPHPSQL
📦 Frameworks
PyTorchScikit-learnDjangoReact JSFlaskOpenCVPySparkAlbumentations
☁️ Cloud & DevOps
DockerKubernetesAzurePrometheusGitAgile / Scrum
🗄️ Databases
MySQLPostgreSQLMongoDBRedis
📊 Data & Analytics
PandasNumPyMatplotlibSeabornTableauJupyter
Proficiency

Skill Levels

// ML & AI Core
PyTorch0%
Python0%
Computer Vision0%
AutoML / NAS0%
Scikit-learn0%
// Frameworks & Tools
Django / Flask0%
React JS0%
Docker / Kubernetes0%
PySpark0%
Azure0%
// Languages
Go0%
C / C++0%
JavaScript0%
SQL0%
PHP0%
Work

Projects Grid

🧠 AutoML · LLM · NAS
LLM-Based Neural Architecture Generation
✦ +11.6% CIFAR-100 · 1,900 architectures

Production AutoML using DeepSeek Coder 7B to generate, validate, and benchmark neural architectures. Introduced Few-Shot Architecture Prompting.

PyTorchDeepSeekAutoML
arxiv.org/abs/2512.24120 →
🏥 Medical AI · Computer Vision
OCT Medical Imaging Segmentation Pipeline
✦ 90%+ val accuracy · Clinical grade

Deep learning pipelines for OCT image segmentation at Universitätsklinikum Würzburg with GPU-accelerated training.

PyTorchOpenCVPySpark
Clinical Research · Confidential
💧 Classification · ML
Water Quality Classification System
✦ 7,000+ samples · Grade A+

End-to-end ML: EDA, feature engineering, multi-model comparison (LR, RF, SVM), Flask deployment with interactive frontend.

PythonScikit-learnFlask
github.com/GoyalAvi →
🌊 Prediction · Environmental ML
Kerala Flood Prediction Pipeline
✦ 3,000+ records · Grade A+ · Ensemble

Supervised learning ensemble (LR, DT, RF, SVM) on Kerala flood dataset with full preprocessing and comparative performance analysis.

PythonScikit-learnPandas
github.com/GoyalAvi →
🌐 Full Stack · Django · React
Petroleum Trade Network Platforms
✦ 7 live websites · Oil & Gas sector

Conference platforms for the Oil & Gas sector: OGAD, Digital Twin, Artificial Lift Summit and more. Built with Django + React.

DjangoReactWordPress
ptnevents.com →
📄 AutoML · LLM · Research
NNGPT: AutoML with Large Language Models
✦ ArXiv Preprint · 2024

Novel approach rethinking traditional neural architecture search using generative capabilities of LLMs for architecture design at scale.

LLMsNASAutoML
arxiv.org/abs/2511.20333 →
CLI

Interactive Terminal

avi@portfolio:~$ — bash
// Welcome to Avi\'s interactive portfolio terminal
// Type a command or click a suggestion below
 
Try: ./about ./skills ./experience ./projects ./contact ./research whoami ls clear
avi@portfolio:~$
Credentials

Certifications & Badges

Google / Udemy
Data Studio A-Z: Data Visualization
2024 📊 Data Analytics

End-to-end mastery of Google Looker Studio for building interactive dashboards, data blending, and advanced visualization techniques for business intelligence.

View Certificate →
Udemy
Data Science With Python: Processing & Visualization
2024 🧠 Data Science

Comprehensive data pipeline design with Python — from raw data ingestion and cleaning to advanced visualization with Matplotlib, Seaborn, and Pandas.

View Certificate →
HR
HackerRank
Python Programming Certificate
2023 ⚡ Programming

Verified proficiency in Python covering data structures, algorithms, OOP, and functional programming paradigms — assessed through hands-on coding challenges.

View Certificate →
Let\'s Connect

Open to Opportunities

Looking for ML engineering, data science, or AI research roles. Let\'s talk about building intelligent systems together.

Available for Opportunities Open to full-time ML / Data Science roles · Würzburg, Germany (or Remote)
Master's completion: Oct 2025
✉ Email 💼 LinkedIn 🐙 GitHub 📱 +49 176 87867118
AI Ask Avi's AI
🤖
Avi's AI Assistant
● Powered by Claude · Ask me anything
👋 Hi! I'm Avi's AI assistant. Ask me anything about his experience, skills, projects, or whether he'd be a great hire — I'm happy to chat! 🚀