/hjhirp

MIT LicenseMIT

Hi there, I'm Harshal Hirpara! 👋

I'm an aspiring 🚀 ML engineer and M.S. Computer Science student @ University of Illinois Chicago 🌆

I love building intelligent systems to solve real-world problems with cutting-edge technology. 💡

🔥 Superpowers Unleashed:

  • Programming Mastery: Python (NumPy, Pandas, Scikit-learn, TensorFlow, PyTorch, Keras), Java, C++, and Scala.
  • Machine Learning Expertise: Deep Learning, Natural Language Processing, Computer Vision, Reinforcement Learning, Gradient Boosting.
  • Data & Cloud Proficiency: Hadoop, Spark, MongoDB, PostgreSQL, AWS (SageMaker, Lambda, EC2), GCP, Azure, Docker, Kubernetes, Terraform.
  • AI & Model Development: Expertise in LLMs, neural networks, and deploying scalable ML models.

🚀 Adventures in AI (Work Experience):

  • Graduate Research Assistant - University of Illinois at Chicago (Aug 2024 - Present)

    • Enhanced user preference-based generation quality for Large Language Models (LLMs) by developing novel methods for alignment.
    • Developed an Offline Reinforcement Learning model to recommend optimal treatment actions for a digital twin of a cancer patient.
    • Tech Stack: Python, PyTorch, RL-Gym.

  • Graduate Researcher - UI Health, Neurology and Rehabilitation Dept. (Oct 2023 - Present)

    • Deployed a scalable ML pipeline reducing EEG signal analysis processing time by 40%.
    • Achieved state-of-the-art detection with 99% accuracy, 88% sensitivity, and 0.43 false positives per hour.
    • Tech Stack: Python, JAX, SciPy, Azure ML Studio.

  • Machine Learning Intern - Cactus Communications (Jan 2023 - Jul 2023)

    • Reduced operational costs by $10K annually by optimizing GPT-3.5 API usage, and improved system reliability by 15%.
    • Created a text extraction and keyword generation tool using Huggingface transformers to aid scientific writing.
    • Tech Stack: Python, PyTorch, AWS (EC2, Inferentia), Docker, Terraform.

📝 Recent Medium Article:

  • Exploring Large Language Models: Concepts, Alignment Techniques, and Practical Implementation 📝
    In this article, I delve into training methods for large language models and alignment techniques, including practical implementations of LoRA, QLoRA, and RLHF. Check it out here.
    For more of my writing, visit my Medium.

🎯 Projects I'm Proud Of:

  • LLM-Grounded Text-In-Image Generation 🖼️: Leveraged Llama 16B to generate conditional text masks within images, utilizing a custom dataset and advanced model optimization techniques.

    • Tech Stack: Python, PyTorch, OpenCV, C++.

  • Ambulatory EEG Signal Analysis 🧠: Reduced EEG signal processing time by 40% with a scalable ML pipeline and achieved 99% accuracy and 88% sensitivity in EEG anomaly detection.

    • Tech Stack: Python, JAX, SciPy, Azure ML Studio.

  • Evaluating LLM Powered AI-Agents 🎮: Developed AI agents generating FAST APIs of 7B and 16B LLMs to navigate complex game environments like blackjack and pathfinding, focusing on algorithm efficiency and bias reduction through few-shot learning and algorithmic adjustments.

    • Tech Stack: Langchain, Python, PyTorch, FastAPI.

  • CS553 Distributed Computing Project - Distributed Algorithms Simulation 🖥️
    This project simulates various distributed computing algorithms implemented in Scala using the Akka framework. The project includes automated simulations, message-passing, shared memory algorithms, and performance visualization with Prometheus and Grafana.

    • Tech Stack: Scala, Akka, Grafana, Prometheus, IntelliJ.

  • SCIPASUMM 📜: Working on an end-to-end research paper summarization pipeline using NLP techniques like Bart-ls.

    • Tech Stack: Python, Bart-ls, Huggingface, PyTorch.

  • Apple Grading Using Computer Vision 🍏: This repository demonstrates how to grade apples using image processing and CNNs. Learn about computer vision, and contribute to agricultural technology. 🌱🤖

    • Tech Stack: Python, OpenCV, Keras, TensorFlow, CNN.

  • Gaze Detection Using Computer Vision 👀: Driver Gaze and Drowsiness Detection with Computer Vision. This project detects driver gaze and drowsiness in real time, enhancing road safety. 📹🤯

    • Tech Stack: Python, OpenCV, MediaPipe.

  • UNET Implementation for Image Segmentation 🌟: Explore the UNET architecture, a powerful tool for image segmentation. Enhance your image segmentation skills through this implementation.

    • Tech Stack: Python, Keras, TensorFlow, UNET.

  • YOLO Implementation for Mars Anomaly Detection 🚀: Discover the secrets of Mars using YOLO to detect anomalies. Explore space exploration and planetary science with this project.

    • Tech Stack: Python, YOLOv3, Keras, OpenCV.

  • Tweet Topic Modelling 📊: Analyze tweet topics using Gensim and Seaborn. Visualize diverse tweet topics with NLP techniques.

    • Tech Stack: Python, Gensim, Seaborn, NLTK, Matplotlib.

  • Emotion Recognition with Face Mask 😷: Created a CNN to classify emotions from masked faces using OpenCV and Keras.

    • Tech Stack: Python, OpenCV, Keras, TensorFlow, CNN.

  • Lithium-Ion Battery 🔋: Built a forecasting model using TensorFlow to estimate battery capacity. Achieved ≤4% error!

    • Tech Stack: Python, TensorFlow, Pandas, NumPy, Time-Series Analysis.

  • Robotic Arm 🤖: Trained a simulated robotic arm to grab objects using reinforcement learning algorithms.

    • Tech Stack: Python, PyTorch, RL-Gym.

  • Plant Disease Classification 🌱: Developed an image classifier using ResNet to identify 38 plant diseases accurately. 🌳

    • Tech Stack: Python, Keras, TensorFlow, ResNet, OpenCV.

There's so much more I'm learning and building as an aspiring ML engineer. Let's connect on LinkedIn! I'm always happy to network with others who are passionate about AI. 😊