** Languages & Web Tech: **

** AI, ML, Deep Learning & Backend: **

** Computer Vision, NLP & Robotics: **

** Data Science & Visualization: **

** Cloud & DevOps: **

** Core Competencies & Specializations: **

• ** 3D Perception & Vision: ** SPVCNN, PointNet++, LiDAR, 2.5D Mapping, GridMap25D, Object Detection
• ** Engineering & Architecture: ** Data Structures, System Design, Software Architecture, REST APIs, Database Design, EDA, Feature Engineering

• Tech Stack: Python, C++17, CUDA, PyTorch, SPVCNN, PyBind11, NumPy, Numba, ROS2, SemanticPOSS
• Overview: Developed a real-time foveated LiDAR perception pipeline that converts raw 3D point clouds into a variable-resolution 2.5D elevation map with semantic, traversability, confidence, and point-density layers. Implemented distance-adaptive spatial representation with 5 cm resolution (0–10 m), 15 cm (10–40 m), and 50 cm (40–100 m) to preserve near-field geometric detail while reducing computational overhead at longer ranges, along with semantic segmentation for drivable terrain, static obstacles, and dynamic objects.
• Performance: Achieved 23.37 ms production-equivalent end-to-end perception latency (~42.79 FPS) after native C++/CUDA and FP16 inference optimizations, reducing pipeline latency by ~75% (from 94.10 ms to 23.37 ms), accelerating 2.5D grid rasterization by 3.88x using C++/PyBind11, and maintaining 52.05% semantic mIoU with zero frame drops across 1,000-frame endurance testing.

• Tech Stack: Python, PyTorch, Transformers, LLMs
• Overview: Developed a from-scratch PyTorch implementation of the Kimi K3 LLM architecture, translating complex research paper specifications into functional code without relying on external reference implementations.
• Performance: Engineered custom neural network components—including Gated Multi-Head Latent Attention (MLA), Attention Residuals (AttnRes), and Stable Latent Mixture-of-Experts (MoE) layers—to optimize context retention and computational efficiency.

• Tech Stack: Python, Scikit-learn, SVM, StandardScaler, Streamlit, Joblib
• Overview: Developed and deployed an end-to-end medical risk diagnostic web application using Support Vector Machine (SVM) and custom CSS Dark Theme UI. Includes diagnostic parameter metrics and downloadable patient reports.
• Performance: Achieved ~77.3% Test Accuracy with standard feature scaling.

• Tech Stack: Python, Scikit-learn, Pandas, NumPy, Jupyter Notebook
• Overview: Built an end-to-end Binary Classification machine learning pipeline using SONAR frequency data to classify underwater objects as either a Rock or a Mine.
• Performance: Achieved ~83.4% Training Accuracy and ~76.1% Test Accuracy using Logistic Regression.

Completed industry-recognized virtual job simulations in Software Engineering, AI, Data Science, Cloud Computing, and Cybersecurity:

• Walmart Global Tech — Advanced Software Engineering
• Deloitte Australia — Data Analytics, Technology & Cybersecurity
• Amazon Web Services (AWS) — Solutions Architecture
• British Airways — Data Science
• Tata — Generative AI Powered Data Analytics

• B.Tech in Computer Science Engineering (Artificial Intelligence)
  University of Lucknow (Expected 2029)

• CIQ Level 7 — Machine Learning Algorithms
• Walmart Global Tech — Advanced Software Engineering (Forage)
• AWS — Solutions Architecture (Forage)
• British Airways — Data Science (Forage)
• Tata — Generative AI Powered Data Analytics (Forage)
• Deloitte Australia — Data Analytics, Technology & Cybersecurity (Forage)
• Successfully completed multiple industry-recognized virtual experience programs from leading global organizations.