Portfolio — 2026
Irfan Firosh
CS & Mathematics at Purdue University. I build intelligent systems at the intersection of AI research, backend engineering, and product. Currently researching LLM reasoning on graph-structured data.
About
Building things
that matter
I’m a third-year CS + Mathematics student at Purdue University (GPA: 3.99) with a passion for building systems that sit at the edge of research and product.
My work spans AI/ML research, full-stack engineering, and cloud infrastructure. I’m most interested in LLM reasoning, graph neural networks, and building products people actually use.
When I’m not in the lab or at a keyboard, I’m probably listening to music, following sports analytics, or chasing the next hackathon.
3.99
GPA
4+
INTERNSHIPS
3+
RESEARCH ROLES
2027
EXPECTED GRAD
LANGUAGES
AI / ML
WEB & APIS
INFRA & CLOUD
Experience
Where I’ve
worked
Software Engineering Intern
Dept. of Computer Science, Purdue/West Lafayette, IN
- ◆Benchmarking LLMs (LLaMA-3, DeepSeek-R1) on target & sentiment detection using zero-shot and few-shot ICL across model scales 3B→70B on HPC infrastructure
- ◆Encoding text-attributed graphs as natural language prompts via graph serialization for graph-enhanced LLM reasoning on political social network datasets
- ◆Designing experiments comparing graph-augmented vs. text-only baselines, quantifying structured reasoning improvements through macro-F1 on imbalanced multi-class targets
Software Engineering & ML Intern
ONOW Enable/Chesterton, IN
- ◆Built Python RAG system with LangChain over 120+ survey databases, reducing manual analysis time by 29% for 15+ field agents
- ◆Containerized REST API backend with Docker + CI/CD, cutting client analysis time by 83% via automated LangChain + OpenAI pipeline
- ◆Built real-time ETL pipeline with Kafka and Redis to stream and cache live survey responses
Software Engineering Intern
Weldon School of Biomedical Engineering/West Lafayette, IN
- ◆Implemented PyTorch nnU-Net for 3D pore connectivity analysis and bone deterioration prediction with NIfTI conversion and Roboflow pipelines
- ◆Automated bone segmentation pipeline in Python using OpenCV + pydicom: 168+ DICOM slices, 1.09M+ voxels with morphological refinement stored in PostgreSQL
- ◆Built graph analytics pipeline in NetworkX modeling 3,162+ nodes and 140k+ edges for bone structure mapping
Software Engineer Intern
Mindster/Bangalore, India
- ◆Developed OCR pipeline + Naive Bayes classifier in Scikit-Learn, reducing manual data entry by 41%
- ◆Built full-stack workflow management system with Laravel + MySQL, boosting team efficiency by 35%
- ◆Launched production Laravel app with Vue.js interface serving 30+ users
Research
Academic work
Research at the intersection of machine learning, graph neural networks, and biomedical engineering.
Graph-Augmented LLM Reasoning for Political Sentiment Analysis
Dept. of Computer Science, Purdue University
Jan 2026 – Present
Benchmarking LLaMA-3 and DeepSeek-R1 on target and sentiment detection in political social networks. Encoding text-attributed graphs as natural language prompts to enable graph-enhanced LLM reasoning. Evaluating improvements in structured reasoning via macro-F1 across imbalanced multi-class targets on HPC.
3D Bone Pore Connectivity Analysis via Graph Neural Networks
Weldon School of Biomedical Engineering, Purdue
Jan 2025 – Present
Leading research in 3D pore connectivity analysis using graph neural networks for bone deterioration prediction. Built a comprehensive medical image processing pipeline with PyTorch nnU-Net, OpenCV, and pydicom processing 168+ DICOM slices with 1.09M+ bone voxels at sub-pixel precision. Graph analytics on 3,162+ nodes and 140k+ edges in NetworkX for bone structure mapping.
LSTM & VAR Models for Healthcare Outcome Prediction
Inogen (Data Science Research)
Aug 2023 – May 2024
Conducted comprehensive analysis of 500k+ health dataset records using Python, R, and SQL. Developed and trained LSTM neural networks and Vector Autoregression models achieving 92% prediction accuracy on validation datasets. Collaborated with medical professionals to translate complex health data into actionable insights.
Selected Work
Things I’ve built
Polymarket Copy Trading Bot
Working on nowAutomated Market Prediction
Automated polymarket copy trading bot that tracks and replicates successful traders.
TokBot
Viral Content Automation
Reddit-to-TikTok pipeline with AI voiceovers, FFmpeg video processing, and serverless deployment.
Live
Right now
● LISTENING TO
● ACTIVE REPOS