"Aspiring ML Engineer and interested in backend dev | Final-Year Undergrad

medium

github

Sandipan Paul  • Peerlist

This project contains implementations and experiments comparing two architectures for classifying cats and dogs: LeNet: A classic, lightweight architecture originally designed for digit recognition. AlexNet : A deeper convolutional network with advanced preprocessing, inspired by architectures trained on ImageNet. Both models were trained on the same Cats vs Dogs dataset for a fair comparison.

CNN Architecture from scratch

A powerful, AI-driven legal document search and understanding tool built for lawyers, law students, and legal researchers. This system leverages vector embeddings and pinecone to help users find, compare, and understand legal cases with ease and accuracy.

LawVector

This project implements a fully connected neural network using only NumPy, without relying on deep learning frameworks like TensorFlow or PyTorch. The model is trained on a classification dataset using forward propagation, backpropagation, and gradient descent. The training process includes performance visualization with real-time accuracy tracking and class prediction distribution plots.

MNIST Digit Classifier and visualization in S

Observa is an end-to-end NLP + MLOps pipeline that demonstrates fine-tuned model inference, real-time monitoring, and experiment tracking.

Observa

This project is a high-performance spam classifier built using an Artificial Neural Network (ANN) with TensorFlow and Keras. The classifier distinguishes between spam and ham (non-spam) messages and achieved an accuracy of 98.19% on the test set. The application is deployed using Streamlit as well as Render offering an interactive web interface for users.

Spam-classifier using ML Models

A web application designed to collect user ratings for songs based on their moods and lay the foundation for predicting songs that match a user's emotional state.

Tunemymood

Frameforge helps creators and teams turn long videos into viral-ready clips — fast.🎬 Auto-generates highlight clips 📊 Provides a viral intelligence score ✍️ Drafts LinkedIn posts from the videoBuilt this to eliminate the pain of manual video repurposing.💎 Coming in V2 (Paid Premium):<ul><li>Higher video limits</li><li>Advanced workflows &amp; deeper insights</li></ul>Join the waitlist for early access to the paid premium features 👇Would love honest feedback from the community.

Frameforge

PaperDigest is an AI-powered research paper summarizer that distills complex scholarly articles into concise, digestible insights—leveraging the Gemini API for primary summarization with fallback support from Gensim and Sumy, and even estimates reading time for efficient consumption and added TTS feature for increased accessibility to the needy users.Features<ul><li>AI-Driven Summarization: Uses the Gemini API as the primary engine for high-quality, context-aware summaries, and seamlessly falls back on Gensim and Sumy when needed to ensure consistent results.</li><li>Multiple Input Formats: Supports both PDF uploads and web URL inputs, automatically extracting and processing text from a variety of sources.</li><li>Estimated Reading Time: Calculates and displays an estimated reading time for the full paper, helping users quickly gauge their time investment.</li><li>Asynchronous Processing: Implements background task management to handle long-running summarization tasks and high concurrency, ensuring a smooth user experience.</li><li>User-Friendly HTTP API &amp; Web Interface: Provides an intuitive API and simple web interface that make integration and usage effortless.</li><li>Download &amp; Share Options: Enables users to download the generated summaries and easily share them via email, social media, or other channels.</li><li>Extensible &amp; Cost-Efficient Architecture: Built with extendability in mind and optimized to run reliably on cost-effective server instances, balancing performance with resource constraints.</li><li>Added TTS(text-to-speech) for users with accessibility issues.</li></ul>

Paperdigest

Accident detection and monitoring system using ESP-32

The Motor and Transformer Calculator project is designed to provide a versatile tool for engineers and enthusiasts to perform calculations related to various types of electric motors (DC, BLDC, PMDC) and transformers. The project is implemented in Python and consists of a command-line interface (CLI) that guides users through the process of selecting a motor type and performing specific calculations.