Hey, I'm Tarun 👋

I build AI systems that solve real-world constraints — from assistive vision to autonomous safety and reinforcement learning optimization. 🚀 Flagship Project 🧠 NAPT — Neural Array for Perceptual Text

An assistive vision system that converts scene text into neural stimulation patterns through a novel glyph-to-electrode mapping.

Designed an end-to-end pipeline: OCR → representation → electrode activation Built multi-stage OCR using EAST + TrOCR for low-visibility environments Developed a geometry-aware glyph mapping to reduce electrode density while preserving perceptual accuracy Validated across diverse fonts, layouts, and noisy conditions

→ A translation framework with no prior implementation reference

🧩 Selected Work 🔹 AI-Enhanced Autonomous Vehicle Safety System Built YOLOv8-based pothole detection with size-aware filtering (avoids unnecessary braking) Designed real-time action pipeline mapping detections → brake signals Integrated vision + LiDAR for robust perception Developed simulation environment for edge-case testing 🔹 Reinforcement Learning — Reward Shaping Study Benchmarked RL algorithms across stochastic environments Improved cumulative reward by 30–40% over baseline (Deep SARSA vs DQN) Analyzed convergence stability and failure modes Studied exploration–exploitation tradeoffs 🔹 Zero Waste Alliance

Data-driven food redistribution platform recognized for innovation and social impact.

📄 Research (In Progress) 🔹 High-Performance CUDA Matrix Multiplication

Currently working on a research-oriented implementation and analysis inspired by “CUDA Matrix Multiplication” by Simon Boehm: https://siboehm.com/articles/22/CUDA-MMM

Reproducing and extending optimization strategies for GPU-efficient matrix multiplication Exploring memory hierarchy usage (shared memory, tiling, coalescing) Analyzing performance trade-offs across kernel designs Positioning work toward a formal paper on efficient GPU computation for ML workloads 🧠 What I Care About Systems that operate under real-world constraints Efficient representations (not just bigger models) Bridging ML + system design + human impact 🛠️ Tech Stack

Languages: Python, C++, SQL ML/AI: PyTorch, TensorFlow, YOLOv8, Scikit-learn Systems & Cloud: GCP, Azure, Data Pipelines, Model Deployment Tools: Git, OpenCV, Visualization (Matplotlib, Tableau, Power BI)

📫 Contact LinkedIn: https://www.linkedin.com/in/tarun-malepati Email: tarunmalepati0907@gmail.com ⚡ Beyond Code

Led operations for a 1,000+ attendee tech fest, scaling participation by 25% YoY through structured execution and outreach.