Welcome to my GitHub! I’m a passionate AI Engineer, AI Researcher, and Computer Vision Specialist focused on planetary image enhancement, LLM-based systems, and intelligent automation.
I love building solutions that bridge the gap between space science, artificial intelligence, and real-world applications.
I completed a research internship at ISRO – Space Applications Centre, and recently worked as an AI Intern at Creole Studios, building production-level AI systems.

• 🔭 Developed GAN, ResNet, and U-Net pipelines for planetary image enhancement using reflection padding, feathered stitching, and perceptual quality metrics.
• 🤖 Built LLM-based systems including an SEO Intelligence Agent using Gemini AI + FastAPI for keyword research, title generation, and content structuring.
• ⚙️ Engineered a Zoho Task Automation System integrating APIs (Zoho, Google Sheets, Gmail) to automate workflows and reduce manual effort.
• 🌍 My work spans Mars terrain analysis, automation systems, and AI-driven content intelligence.
• 📚 Continuously exploring Generative AI, Agentic AI, Backend Systems, and Scalable AI Architectures.
• 💬 Ask me about: Deep Learning, Generative AI, LLMs, FastAPI, Automation, Computer Vision, or Space-Tech + AI!
• 📫 How to reach me: LinkedIn | Email

AI & LLM Systems: Gemini AI, Generative AI, LLMs, Prompt Engineering, Agentic AI, RAG
Backend & APIs: FastAPI, AsyncIO, REST APIs, OAuth 2.0, Streaming APIs
Image Enhancement Models: CNN, U-Net, GAN (Pix2Pix, CycleGAN), ResNet, Reflection Padding, Feathered Stitching
Computer Vision & Preprocessing: Satellite Imagery, Crater Detection, MSR, CLAHE, Adaptive Gamma, SSR, Mars Image Enhancement
Metrics (Custom NumPy): BRISQUE, NIQE, PIQE, Entropy, SNR, HVS
Cloud & Compute: NVIDIA DGX, Google Colab GPU, ISRO-HPC

Languages : Python, C, HTML, CSS, Git AI/ML Frameworks : TensorFlow, Keras, PyTorch, OpenCV, Langchain, RAG Concepts : Deep Learning, CNNs, GANs, Computer Vision, Generative AI, LLMs, Prompt Engineering, AI System Design
Backend & Systems : FastAPI, API Integration, Automation Systems, Async Programming, Scalable Architectures
Tools : NumPy, Pandas, Matplotlib, Git, Power BI, Excel, Jupyter, Google Colab, VS Code, Canva
Other : UI Designing, Data Analytics, Workflow Automation
Soft Skills : Analytical Thinking, Problem Solving, Research Mindset, Team Collaboration, Time Management

• 🚀 Building AI-powered automation systems
• 🤖 Developing LLM-based intelligent applications
• ⚡ Designing scalable backend architectures for AI systems
• 🌍 Solving real-world problems using AI + data-driven approaches

• Developed a deep U-Net model with skip connections and attention layers
• Enhanced satellite terrain features with high contextual accuracy across patch boundaries

• Developed and documented a complete planetary image enhancement pipeline as part of a research paper during ISRO internship
• We have evaluated our proposed methods using Indian Mars Colour Camera (MCC) images and compared with other state of the art image enhancement techniques. It involves Adaptive Gamma Correction (AGC) to correct brightness, Contrast Limited Adaptive Histogram Equalization (CLAHE) to locally enhance contrast, noise removal via median filtering, sharpening fine details through unsharp masking, and edge-preserving smoothing by bilateral filtering. In addition, we assess the improved performance of Mars Colour
• Demonstrated results on Martian terrain with quantitative and visual comparisons

• Designed a Pix2Pix GAN to enhance Martian surface images with reflection padding and Gaussian feather stitching
• Improved perceptual quality without introducing boundary artifacts
• Evaluated using BRISQUE, NIQE, and entropy metrics

• Created a patch-level deep learning pipeline for large planetary images
• Implemented seamless stitching, patch line removal, and adaptive enhancement strategies

• Implemented a ResNet-based enhancement network for improving fine surface details in planetary imagery
• Compared performance with U-Net and GAN pipelines for various patch sizes
• Integrated reflection padding and evaluated using no-reference IQA metrics
• Achieved superior enhancement in low-texture crater regions and edge fidelity

• Built a complete image classification pipeline using CIFAR-10 dataset
• Developed a baseline CNN and advanced to MobileNetV2 transfer learning
• Implemented training, evaluation, and real-time inference scripts
• Visualized results with confusion matrices, training history, and accuracy plots
• Designed as an educational and research-friendly framework for rapid AI prototyping

• Multi-Language Support – Offline Speech Recognition for English, Hindi & Gujarati using optimized language models.
• Lightweight Execution – Core files like offline_stt.py, offline_stt_hi.py, and offline_stt_gu.py handle speech-to-text efficiently without heavy dependencies.
• Configurable & Scalable – settings.json allows easy customization, and the project supports modular expansion for additional languages.
• Testing Made Easy – test_stt.py enables quick testing & validation of all models for smooth execution.
• External Model Hosting – Heavy models stored on Google Drive for easy download.

• End-to-End Automation - Fully scheduled n8n workflow that automates trending topic research without manual execution.
• Dynamic Keyword Selection - Rotational keyword strategy ensures diversified and unbiased trend discovery across executions.
• Multi-Platform Data Integration - Aggregates trending signals from YouTube Data API v3 and NewsAPI with fault-tolerant design.
• AI-Driven Content Ideation - LLM-oriented analysis pipeline designed to synthesize cross-platform trends and generate strategic content ideas.
• Structured Output Storage - Automatically persists insights into Google Sheets using timestamped, structured records for downstream use.

• 3-Step Intelligent Workflow - Structured pipeline that sequentially performs keyword extraction, title generation, and outline creation with controlled user interaction.
• Search Engine Simulation - Uses LLM-based reasoning to replicate real Google search behavior and identify top-ranking pages with high SEO relevance.
• Advanced Keyword Intelligence - Extracts and deduplicates high-ranking SEO keywords based on content signals from authoritative sources.
• SEO-Optimized Title Generation - Produces high-conversion blog titles aligned with search intent, keyword placement, and ranking patterns.
• Dynamic Outline Generation - Builds context-aware blog structures with variable H2/H3 hierarchy based on topic depth and semantic coverage.
• User-Guided Content Flow - Enables keyword and title selection to improve relevance and downstream content quality.
• Robust URL Validation Pipeline - Implements filtering including invalid links, soft 404 detection, and topic relevance validation.
• Real-Time Streaming Architecture - Uses FastAPI streaming responses to deliver progressive outputs with step-wise feedback.
• Fault-Tolerant Generation System - Includes retry logic and strict JSON parsing to handle inconsistent LLM outputs.
• Modular Backend Design - Clean separation of agent logic, LLM client, and utilities for scalability and maintainability.

• End-to-End Task Automation - Built a backend system to automate task creation, tracking, and workflow execution across Zoho Projects.
• Multi-Platform Integration - Seamlessly integrates Zoho Projects, Google Sheets, and Gmail for synchronized data flow and communication.
• Bulk Task Processing Engine - Enables large-scale task creation from structured Google Sheets data with optimized batch operations.
• Smart User Mapping System - Implements persistent user ID management to ensure accurate task assignment and tracking.
• Automated Notification System - Syncs task updates with Gmail to trigger automated email reminders and communication flows.
• Hierarchical Task Management - Supports parent-child task structures for better organization and project scalability.
• OAuth 2.0 Authentication Handling - Secure API integration with automatic token refresh for uninterrupted system operation.
• Performance Optimization Pipeline - Reduced execution time from ~2 minutes to ~5 seconds using batch API calls and efficient data handling.
• Error Handling & Data Validation - Ensures reliability through structured validation, exception handling, and fault tolerance.
• Scalable Backend Architecture - Built using FastAPI with modular design for maintainability and production-level scalability.

• 🎓 Research Intern – ISRO, Space Applications Centre, Ahmedabad
• 🌍 Developed and author of a research paper on planetary image enhancement pipelines for Earth and interplanetary missions
• 🏅 Built fully custom no-reference image quality metrics (BRISQUE, NIQE, PIQE, etc.) without third-party libraries on ISRO DGX cluster
• 🚀 Contributor to advanced AI solutions for satellite change detection and image restoration in ISRO
• AI Tools in the Workplace Certificate : Covered Prompt Engineering, ChatGPT - June 2025
• Deloitte Data Analysis Virtual Internship (Forage) – June 2025
• Predictive Modeling : Forecast Like a Pro - July 2025
• Excel Using AI Workshop - July 2025
• Work Ready Certificate by BEEP and Shark Tank - July 2025
• AI Dashboards using Microsoft Power BI - July 2025
• Machine Learning with Python : k-Means Clustering by LinkedIn Learning - July 2025
• 5 Days Python for Data Science Bootcamp - July 2025
• Freedom withh AI Masterclass - August 2025
• BUILD AN AI THAT SEES : IMAGE RECOGNITION - August 2025
• IEEE Student Branch Member – 2022 to 2023 Video and Image Editing Contributor – IEEE Student Branch

• 🔗 LinkedIn
• 📬 Email

Thanks for visiting my profile!
Let’s build intelligent solutions that explore the universe. 🚀

⭐️ From Shivam Raval