Neural Nets

Welcome to my blog! I'm Rahul, your guide on an exciting journey through the worlds of Python, deep learning, and cybersecurity. Whether you're here to learn, explore, or deepen your understanding, you're in the right place. Together, we'll navigate complex concepts and make them accessible — no prior expertise required! 🌟🤗

Latest Posts

Projects

Neural Nets

Architected and deployed a scalable, cloud-native blog infrastructure on Google Cloud, utilizing Cloud Run, Cloud Storage, and Cloud CDN for secure and efficient content delivery. Integrated BigQuery for advanced analytics and Cloud Logging for real-time monitoring. Leveraged Cloud Storage as a DataSink to optimize performance tracking and reporting. Additionally, implemented the Cloud Text-to-Speech API to convert articles into audio format, enhancing accessibility and user engagement. This solution ensures high availability, operational efficiency, and actionable, data-driven insights.

Project Graphil

Designed an interactive platform to simplify technical concepts through dynamic visualizations. It offers interactive graphs and diagrams on topics like Python, Linux, Networking, Google Cloud, and AI. Built with React, Graphil serves as a powerful tool for learners and educators to understand complex topics through visual learning.

Mission Cipher

Developed and deployed a Retrieval-Augmented Generation (RAG) web app on Compute Engine, efficiently delivering contextually relevant information about the Mission Impossible film series. The app integrates document retrieval, semantic embeddings, and generative AI to match queries with accurate content using cosine similarity. Optimized backend performance with Gunicorn, configured as a WSGI server, and served through Nginx acting as a reverse proxy. Communication between Nginx and Gunicorn is handled via a Unix socket, ensuring a high-performance, secure connection for real-time queries. Necessary ports have been opened for ingress traffic to facilitate smooth communication and access.

Gemma 2B: Instruction Tuning with LoRA

Fine-tuned Google’s Gemma 2B model on a 4,000-sample subset of the Alpaca dataset using Low-Rank Adaptation (LoRA) to enhance instruction adherence and elevate response fidelity. Model performance was evaluated using perplexity scores, token-level accuracy analysis, and qualitative side-by-side comparisons, showcasing significant improvements in contextual understanding and output relevance.

CloudNet Analytics

Built a high-performance, security-hardened distributed log aggregation and analytics system on Google Cloud. A custom Virtual Private Cloud (VPC) with meticulously segmented web (x.y.1.0/24), app (x.y.2.0/24), and processor (x.y.3.0/24) subnets, each hosting a dedicated virtual machine (VM) and defined by specific CIDR blocks, ensures robust network isolation and adherence to zero-trust security principles. Firewall rules enforce strict access control, allowing only preconfigured IPs. A streamlined log ingestion pipeline securely transfers logs over SSH, processes data using Python, and integrates with Cloud Storage and Pub/Sub. Cloud Functions (1st gen.), triggered by Pub/Sub events, automate structured data ingestion into BigQuery. A Flask-based API provides controlled external access to processed logs and analytics

Compliance Guide

Created a detailed compliance guide for a fictional digital business, focusing on essential regulatory frameworks and cybersecurity standards. The guide provides actionable recommendations for areas like data protection, PCI compliance, risk management, and employee training, ensuring businesses align their operations with industry regulations to maintain a strong security posture.