Applied AI/ML Developer → Architecting GenAI (RAG/LLMs) & Computer Vision Systems | Deep Data Analysis

github

projects

youtube

Tuhin Kumar Singha Roy  • Peerlist

QR Forge — Free Custom QR Code GeneratorQR Forge is a comprehensive, free online QR code generator built as a Streamlit web application. It offers advanced features beyond basic QR code creation, including three distinct modes: trackable dynamic QR codes with expiry-based redirects, artistic QR codes that visually mimic an uploaded image, and permanent QR codes that embed the destination URL directly without server dependency.The application is designed for production readiness, incorporating SEO meta injection, structured JSON-LD data, scan analytics with GitHub auto-push capabilities, and a polished user interface.Key Features: Three QR Modes: Dynamic (trackable redirect), Artistic (image-mapped), and Permanent (serverless).Customization: Options for custom colors (foreground/background), logo overlay, and background image blending.Error Correction: Adjustable error correction levels (L, M, Q, H).Dynamic Features: Expiry timer for redirects (1 minute to 7 days) and scan analytics.Artistic Mode: Image upload for visually unique QR codes with adjustable saturation and background styles.Permanent Mode: Direct URL embedding for indefinite QR code functionality.Data Management: Analytics stored in analytics.json and automatically pushed to GitHub.SEO Optimization: Runtime injection of meta tags, Open Graph, Twitter Cards, and JSON-LD structured data.Development Stack: Built with Python, Streamlit, segno, and Pillow.

QR-forge 

Architected and implemented an AI-driven trading system using Time Series Analysis and TensorFlow for trend prediction. Built a proprietary backtester for Continuous Training (CT) evaluation and integrated a Google Sheets logging system for transaction auditability. Delivered instant trade alerts via Telegram Bot API.

AI Trading System

Built an intelligent web-based application that classifies emails as Spam or not spam using the Google Gemini LLM API. The system analyzes email content based on urgency, suspicious links, grammar, and more, returning a structured response with: ✅Classification ✅Confidence Score ✅Reasoning ✅Spam Indicators ✅Risk Level. The app also features real-time classification, historical tracking, data export, and a polished UI using Streamlit and custom CSS. Prompt engineering was used to guide Gemini's output in structured JSON format.

AI-Powered Email Classifier(Spam or Not Spam)

Built an enterprise-grade Generative AI agent for academic document analysis and secure Q&amp;A. Utilized PyMuPDF, Pandas, NumPy for multi-modal processing to extract text, tables, and figures, and integrated ArXiv API for seamless search and analysis of academic papers.

Askademia

This project involves the development of two machine learning models to recommend the most suitable crops and fertilizers based on soil and environmental data. Using CSV datasets containing parameters like nitrogen (N), phosphorus (P), potassium (K), pH, temperature, humidity, and rainfall, Random Forest classifiers were trained and saved as deployable .pkl files (Crop_Recc_model.pkl and Fertilizer_Recc_model.pkl). These models were deployed using Streamlit and Flask for Prediction.

Crop and Fertilizer Recommendation System

Engineered a robust, production-ready Retrieval-Augmented Generation (RAG) system for deep context-aware analysis of large-scale, multi-document PDF sets. Leveraged LangChain and Gemini API to enable persistent conversation history, support complex queries, and deployed via interactive Streamlit interface. Optimized for real-world document search and analytics.

Deep Research Assistant (RAG Architecture)

Tools: Mistral AI, Flask, Streamlit, HTML/CSS, JS. Farm Friend is an intelligent chatbot solution designed to assist farmers by providing real-time crop and fertilizer recommendations. Farmers input details such as soil type, temperature, moisture, and location, which are processed and sent as a prompt to Mistral AI. Based on this information, the AI generates personalized suggestions for the most suitable crops and fertilizers, including location-specific recommendations. The chatbot is designed to bridge the gap between AI advancements and agricultural practices, empowering farmers with smarter decision-making tools.

Farm Friend – AI-Powered Agricultural Chatbot

Developed an internal recruitment web application to streamline candidate filtering and summary generation using the Gemini API. Designed a customizable scoring system that classified candidate relevance with 95% matching accuracy, enhancing hiring workflows and decision-making.

Hiring Nexus - AI Hiring Helper

Mistral AI Chatbot is an advanced conversational AI application designed to provide intelligent and interactive responses using the Mistral API. Built with Flask, Python, and JavaScript, the chatbot delivers a seamless user experience with a futuristic Jarvis UI featuring neon effects and glitch animations. The chatbot maintains chat history, supports real-time messaging, and integrates local storage for saving conversations. The backend, powered by Flask, handles API calls efficiently, ensuring quick and accurate responses. The front-end interface is designed with HTML, CSS, and JavaScript, providing an intuitive and engaging user experience. This project showcases expertise in LLM (Large Language Models), API integration, frontend development, and UI/UX design, making it a valuable contribution to AI-powered chatbot applications.

Mistral AI Chatbot

I have Successfully developed a Convolutional Neural Network (CNN)-based deep learning model as part of my internship, focusing on precision agriculture. The project aims to detect and classify plant diseases from high-resolution images of crop leaves, including apples, cherries, grapes, and corn. The model is being designed to accurately distinguish between healthy and diseased leaves while predicting specific types of diseases. My ongoing responsibilities include: - Collecting and preprocessing a diverse dataset of leaf images. - Designing, training, and optimizing the CNN architecture to achieve high accuracy and robustness. - Applying techniques such as data augmentation and transfer learning to improve model performance. - Evaluating the model using metrics like accuracy, precision, recall, and F1-score to ensure reliability. - Exploring deployment strategies for real-time predictions to facilitate early disease detection and effective management. - Deploy the project on GitHub and using Streamlit. This experience has allowed me to enhance my expertise in computer vision, deep learning, and agricultural technology, demonstrating my ability to tackle real-world challenges with innovative machine learning solutions.