Projects That Matter
Six projects across AI, computer vision, full-stack web, and data analytics — each solving a real problem with precision-engineered code. Scroll through to explore each one.
DashNexus
An AI-powered analytics platform that transforms raw CSV/Excel files into interactive dashboards in minutes. Upload your data — DashNexus automatically cleans it, runs feature engineering, generates AI insights, and builds charts. Explore results through a conversational chat interface, voice queries, or direct editing. Built as a full production SaaS with subscription monetization.
- Automated data cleaning, feature engineering & profiling pipeline
- AI-powered chart generation from natural language prompts
- Conversational + voice query interface for dashboards
- Subscription monetisation — production-ready SaaS architecture
Gaandiv
An AI-powered knowledge-base chatbot using Retrieval-Augmented Generation (RAG) to answer domain-specific queries accurately. Built with a custom retrieval pipeline: embedding-based search with a TF-IDF fallback, and Llama 3.1 via Cloudflare Workers AI for response generation. Features multilingual support, session chat history, admin knowledge management, and a glassmorphism UI.
- Custom embedding retrieval pipeline with TF-IDF fallback
- PDF & HTML document processing for knowledge ingestion
- Multilingual chat + session-based history + admin panel
- Glassmorphism UI with markdown rendering & code highlight
Emotion
Detect
A Django + TensorFlow web application that classifies 7 human emotions in real-time from live webcam feeds or uploaded images. The CNN was trained on the FER2013 dataset using data augmentation and regularisation for improved accuracy. Features global model caching for high-speed inference and a modern glassmorphism UI with animated gradient backgrounds.
- 7-class recognition: Angry, Disgust, Fear, Happy, Sad, Surprise, Neutral
- Real-time webcam feed with instant visual feedback overlay
- Global model caching for high-performance inference
- Animated gradient glassmorphism frontend design
Trip—Z
A full-stack Django travel assistant with an AI-driven chat UI that saves sessions, integrates an LLM for personalised travel recommendations, supports PDF export of complete itineraries, and manages subscriptions with Razorpay payments. Built with a layered service architecture: prompt_builder, gemini_service, pdf_service, and razorpay_service.
- Persistent chat sessions with LLM-powered travel advice
- Prompt builder + Gemini service for itinerary generation
- PDF export of complete trip plans via pdf_service
- Subscription plans with live Razorpay payment integration
Object
Detect
A YOLOv8-powered real-time object detection system supporting multiple input sources: live webcam, video files, static images, and UAV dataset batches. Features trajectory tracking with a history deque, direction prediction (Up/Down/Left/Right), live object counting, and a real-time analytics dashboard — all running in a multi-threaded architecture that keeps video playback smooth.
- Webcam, video, image & UAV dataset batch processing pipelines
- Trajectory tracking with history deque & direction prediction
- Real-time object counting with time-series chart visualisation
- Multi-threaded live plotting without blocking the video feed
URL
Maliciousness
Detect
A decoupled Django + ML microservice system for real-time URL maliciousness classification. The Django frontend collects user-submitted URLs and displays detection history, while a separate Python ML service handles feature extraction, model inference (scikit-learn), and exposes a REST API. Results — malicious or benign, with confidence — are stored and visualised.
- Decoupled architecture: Django UI + standalone ML service communicating via HTTP
- Feature extraction pipeline from raw URLs with trained scikit-learn classifier
- Malicious / Benign output with confidence scoring per request
- Detection history dashboard stored in SQLite with visual summary