Kennedy Kiragu Maina

Enhancing RAG Systems with Fine-Tuned Models

This project focused on creating a Retrieval-Augmented Generation (RAG) system by utilizing a Google’s fine-tuned Gemini model. The goal was to enhance prediction accuracy and system efficiency in analyzing sensor data.

Key Contributions:

• Implemented local TensorFlow.js models for real-time predictions.
• Developed advanced preprocessing techniques for sensor data normalization, sequence generation, and 3D tensor reshaping.
• Upgraded to Gemini 2.0 Flash Exp model for enhanced query understanding, data retrieval, and summarization.

Technologies Used: Next.JS, NestJs, TypeScript, TensorFlow.js, ShadCn UI, DaisyUI, Google Gemini Model, TensorFlow Layers API

Gitlab Repository

Code Snippet (Data Preprocessing):

private preprocessData(data: any) {
    const normalizedData = data.map((row) => ({
        ultrasonic: row.ultrasonic / 100,
        seismic: row.seismic / 100,
        tilt: row.tilt / 100,
        load: row.load / 100,
        pressure: row.pressure / 100,
        sound: row.sound / 100,
    }));
}

                

Real-Time Delivery Tracking System

A React-based application using OpenLayers for real-time visualization of delivery trucks and their destinations.

Key Features:

• Live truck location tracking on an interactive map.
• Shop markers indicating delivery destinations.
• Potential for route optimization and customer notifications.

Technologies Used: Next.js, TypeScript, OpenLayers, Vercel

Live Demo: Truck Tracker

GitHub Repository

Full-Stack Todo List Application

A user-friendly Todo List application with Google Sign-In integration for seamless authentication and real-time task management.

Key Features:

• Google Sign-In and Firebase Authentication.
• CRUD operations for managing tasks.
• Responsive interface built with Tailwind CSS.

Technologies Used: Next.js, TypeScript, Tailwind CSS, Node.js, Express.js, Firebase Firestore, Vercel

Live Demo: Todo List App

Github Repository

Dating App API

Built a robust API for a dating application with scalable architecture, real-time database management, and containerized deployment.

Key Features:

• NestJS framework for modular and scalable design.
• Prisma ORM for PostgreSQL database operations.
• Docker for seamless development and deployment.
• Swagger for interactive API documentation.

Technologies Used: NestJS, Prisma, Docker, PostgreSQL, Swagger

GitHub Repository

Aviator Game Reimagined

Developed an interactive betting game featuring real-time mechanics and engaging user interactions.

Key Features:

• Four simultaneous game sessions where players bet on balloons.
• Real-time communication powered by WebSockets.
• Intuitive user interface for a seamless betting experience.

Technologies Used: Django, Celery, PostgreSQL, Redis, PhaserJS, HTML, CSS, JavaScript

GitHub Repository