Hi there, I'm G Santosh Kumar 👋

Projects

1. Boxsand Coder Agent

An Intelligent AI Developer Workspace A VS Code-style browser IDE powered by Pydantic-AI and the Model Context Protocol (MCP). It functions as an autonomous coding partner capable of managing file systems, executing git operations, and generating code in real-time.

⚡ Tech Stack & Architecture:

• Core AI: Python 3.11, Pydantic-AI, Google Gemini Models.
• Backend: FastAPI, MCP (Model Context Protocol) Server, Cloud Run.
• Frontend: React 19, Vite, Monaco Editor, XTerm.js.
• Architecture: Agent-based Microservices with tool registration and stateful session management.

---

2. Occupancy Analytics & People Tracking

Enterprise MLOps & Computer Vision Pipeline A comprehensive MLOps-IoT platform designed for automated video surveillance. It automates the training, deployment, and maintenance of computer vision models to track occupancy, demographics (gender, age), and workplace safety in real-time.

⚡ Tech Stack & Architecture:

• ML Core: PyTorch 2.4, Ultralytics YOLOv8, OpenCV.
• Orchestration: Kubeflow Pipelines (KFP 2.0), Vertex AI Training.
• Data: BigQuery, Firestore, Google Cloud Storage.
• Architecture: Event-Driven MLOps Pipeline with automated retraining triggers and edge-compatible deployment.

---

3. Podcraftor

Text-to-Podcast Automation Engine A full-stack application that converts plain text into complete podcast episodes. Utilizes Google TTS with SSML support to generate natural, human-like audio, automating podcast production end-to-end.

⚡ Tech Stack & Architecture:

• Backend: Python 3.12, FastAPI, LangChain (LLM Orchestration).
• Audio Processing: FFMpeg, Pydub, Google Cloud TTS (SSML).
• AI Models: Google Vertex AI, Gemini Pro.
• Architecture: Service-oriented architecture with Factory patterns for TTS providers and Strategy patterns for content generation.

---

4. Audiobook Generator

Intelligent Document-to-Audio System An intelligent full-stack solution that processes PDFs and ePUB formats to autonomously structure chapters and generate high-quality audiobooks using Google TTS — surpassing traditional audiobook features offered by platforms like ElevenLabs.

⚡ Tech Stack & Architecture:

• Backend: FastAPI, Pydantic, SpaCy (NLP), Ebooklib.
• Frontend: Next.js 14, React 18, Firebase Auth.
• Cloud: Cloud Run, Cloud Storage, Google Cloud Vision API.
• Architecture: Async-first microservices with background task scheduling (APScheduler) and stream processing.

---

5. Medical Docs Analyzer

Healthcare Fraud Detection System A smart healthcare document analysis system capable of interpreting diverse medical documents including handwritten prescriptions and discharge summaries. It extracts key KPIs, detects fraud, verifies document legitimacy, and generates contextual follow-up questions.

⚡ Tech Stack & Architecture:

• Core: Next.js 14, TypeScript, PDFKit, Sharp.
• AI: Gemini 2.0 Flash (Multimodal Analysis).
• Infrastructure: Cloud Run (Serverless), Cloud Build.
• Architecture: Serverless Multi-Step AI Pipeline (Classification → Analysis → Fraud Detection).

---

6. Pixora Studio

AI-Powered Image and Video Generation Studio A "Photoshop Agent" and Video Generation studio that unifies multiple generative models into a single creative workflow. It handles complex media operations like video trimming and composition directly in the browser.

⚡ Tech Stack & Architecture:

• Frontend: Next.js 14, React 18, MediaRecorder API.
• AI Models: Gemini 1.5 Pro, Google Imagen 3, Google Veo (Video Generation).
• Architecture: Multi-Model AI Studio pattern with hybrid client-side video processing and server-side AI generation.

---

7. Live Voice Agent

Real-time 3D AI Interaction An Intelligent live voice agent using Gemini-Live-Voice model. The Agent is integrated with function calling and tool use to take action on behalf of the user, visualized with a reactive 3D avatar.

⚡ Tech Stack & Architecture:

• Core: TypeScript, React, Three.js (WebGL), Web Audio API.
• AI: Gemini Live API (Real-time Streaming).
• Architecture: Client-side audio processing pipeline with bidirectional WebSocket communication.

Video	Screenshot
AI Voice bot

---

8. Gifinity

AI Sprite Sheet & GIF Generator A fun App that generates sprite sheet images using Gemini Nano/Pro models and converts them to animated GIFs client-side.

⚡ Tech Stack & Architecture:

• Core: Next.js, HTML5 Canvas API, Gifshot.js.
• AI: Gemini Pro Vision, Gemini Nano.
• Architecture: Client-Side Animation Pipeline where heavy frame processing is offloaded to the browser via Canvas API.

---

9. Podcast Idea Generator

Trend-to-Content Intelligence A sophisticated API-driven application designed to bridge the gap between raw trend data and actionable podcast content. It intelligently combines insights from global Google Trends and internal podcast analytics.

⚡ Tech Stack & Architecture:

• Backend: Python 3.12, FastAPI, Pydantic.
• Data: Google BigQuery (Trend Analysis), Elasticsearch 8.x (Analytics).
• Architecture: Async Microservices orchestration leveraging BigQuery for large-scale trend aggregation.

---

10. Sentiment Analyzer

Social Media Campaign Tracker A comprehensive social media analytics tool that scrapes user comments from platforms like Instagram, YouTube, and Facebook to analyze sentiment trends and track influencer campaign effectiveness.

⚡ Tech Stack & Architecture:

• Core: Next.js 14, React 18, TailwindCSS.
• AI: Gemini 1.5 Flash (Sentiment Classification).
• Infrastructure: Google Cloud Run, Cloud Build (CI/CD).
• Architecture: Serverless Microservices pattern deployed via Cloud Build.