# no bugs here, just features
parina = {
"currently" : "exploring, building & growing in the AI space 🌱",
"learning" : ["LLMs", "RAG pipelines", "Computer Vision"],
"fun_fact" : "I debug better with iced coffee in hand ☕",
"looking_for" : "seeking AI/ML and software engineering internships where I can build scalable, data-driven systems",
}
| skill | tools |
|---|---|
| 🧠 Deep Learning | TensorFlow · Keras |
| 🎯 Computer Vision | ResNet · CNNs · image pipelines |
| 🤖 ML Fundamentals | scikit-learn · NumPy · Pandas |
| 🐍 Python | always Python |
| ☁️ Cloud & DB | AWS · Firebase · MySQL |
| 🌐 A lil web | HTML · CSS · JavaScript · React |
ResNet50 · TensorFlow · Keras · Python · OpenCV
Computer vision model that classifies driver behaviour into 10 categories — safe driving, texting, phone calls, drinking, and more — using the State Farm dataset (22,424 images). Built on ResNet50 with residual learning blocks, Leave-One-Group-Out cross-validation to prevent subject-level data leakage, and full bias-variance analysis. Achieved 86.95% training accuracy over 10 epochs.
ASP.NET Core · C# · SQL Server · Gemini API · Chart.js
AI-powered workforce intelligence platform built during an internship at Surtel Technologies (SAP Silver Partner). Sits on top of enterprise time-tracking data and makes it intelligent — managers can query team productivity in plain English, get Gemini-generated weekly summaries, and receive smart alerts for overload risk, meeting overhead, and sprint health. Uses a prompt chaining pipeline where Gemini first detects anomalies, then generates recommendations from live SQL data.
