ATUL SHUKLA atleekumaar

Key Skills & Technologies

** 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 3.12, FastAPI, React 19, TypeScript, Tailwind CSS, Scikit-learn, Apache Parquet, Groq LPU, Twilio API, Docker
Overview: Engineered an enterprise-grade, localized business operating system and conversational AI assistant for retail merchants across India. Built real-time financial command centers, hyperlocal peer benchmarking across 6 operational metrics (Repeat Rate, Basket Size, Failure Rate, Growth) with hierarchical fallback, an interactive offline cash ledger, and an automated decision engine for inventory restock alerts, customer churn mitigation, and cross-sell bundling with auditable data evidence.
Performance: Implemented a 100% zero-hallucination multilingual AI Copilot (Hindi, Hinglish, English) with ultra-fast Groq LPU inference (llama-3.3-70b / qwen3.8), achieving < 15 ms P95 API response times, 2.08s end-to-end data pipeline throughput across 10,000+ transactions, 98/100 Lighthouse performance score, and 100% pass rate across 173 automated test suites.

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, FastAPI, Grad-CAM, HTML/CSS/JS
Overview: Developed a production-quality clinical AI radiology workstation to classify axial brain MRI images into four categories (Glioma, Meningioma, Pituitary, No Tumor) using a custom PyTorch Convolutional Neural Network. Integrated real-time Grad-CAM visual explainability, an interactive DICOM/PACS-style viewer, and automated PDF clinical research report generation.
Performance: Achieved 94.56% test accuracy, a 0.974 AUC-ROC score, and 96.2% diagnostic sensitivity across 1,600 held-out clinical test slices. Delivered real-time forward-pass execution with ~35 ms inference latency for instant morphological attention visualization.

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.

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

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.